#114 – Russ Tedrake: Underactuated Robotics, Control, Dynamics and Touch

[0] The following is a conversation with Russ Tadrick, a roboticist and professor at MIT and vice president of robotics research at Toyota Research Institute or TRI.

[1] He works on control of robots in interesting, complicated, underactuated, stochastic, difficult -to -model situations.

[2] He's a great teacher and a great person, one of my favorites at MIT.

[3] We'll get into a lot of topics in this conversation from his time leading MIT.

[4] DAPRA Robotics Challenge team to the awesome fact that he often runs close to a marathon a day to and from work barefoot.

[5] For a world -class roboticist interested in elegant efficient control of underactuary dynamical systems like the human body, this fact makes Russ one of the most fascinating people I know.

[6] Quick summary of the ads, three sponsors.

[7] Magic Spooned Serial, BetterHelp, and ExpressVPN.

[8] Please consider supporting this podcast by going to magic spoon .com slash lex and using code lex at checkout, going to betterhelp .com slash lex and signing up at expressvpn .com slash lex pod.

[9] Click the links in the description, buy the stuff, get the discount.

[10] It really is the best way to support this podcast.

[11] If you enjoy this thing, subscribe on YouTube, review it with five stars on Apple podcast, support it on Patreon, or connect with me on Twitter at Lex.

[12] Friedman.

[13] As usual, I'll do a few minutes of ads now and never any ads in the middle that can break the flow of the conversation.

[14] This episode is supported by Magic Spoon, low -carb, keto -friendly cereal.

[15] I've been on a mix of keto or carnivore diet for a very long time now.

[16] That means eating very little carbs.

[17] I used to love cereal.

[18] Obviously, most of crazy amounts of sugar, which is terrible for you, so I quit years ago.

[19] But Magic Spoon is a totally new thing.

[20] zero sugar, 11 grams of protein, and only three net grams of carbs.

[21] It tastes delicious.

[22] It has a bunch of flavors.

[23] They're all good, but if you know what's good for you, you'll go with cocoa, my favorite flavor and the flavor of champions.

[24] Click themagicspoon .com slash Lex.

[25] Link in the description.

[26] Use code Lex at checkout to get the discount and to let them know I sent you.

[27] So buy all of their cereal.

[28] It's delicious and good for you.

[29] You won't regret it.

[30] The show is also sponsored by BetterHelp, spelled H -E -L -P -Help.

[31] Check it out at BetterHelp .com slash Lex.

[32] They figure out what you need and match you with a licensed professional therapist in under 48 hours.

[33] It's not a crisis line.

[34] It's not self -help.

[35] It is professional counseling done securely online.

[36] As you may know, I'm a bit from the David Goggins line of creatures, and so have some demons to contend with, usually on long runs or all -nighters, of self -doubt.

[37] I think suffering is essential for creation, but you can suffer beautifully in a way that doesn't destroy you.

[38] For most people, I think a good therapist can help in this, so it's the least worth a try.

[39] Check out the reviews.

[40] They're all good.

[41] It's easy, private, affordable, available worldwide.

[42] You can communicate by text anytime and schedule weekly audio and video sessions.

[43] Check it out at betterhelp .com slash Lex.

[44] This show is also sponsored by ExpressVPN.

[45] Get it at ExpressVPN .com slash LexPod to get a discount and to support this podcast.

[46] Have you ever watched The Office?

[47] If you have, you probably know it's based on a UK series also called The Office.

[48] Not to steer up trouble, but I personally think the British version is actually more brilliant than the American one.

[49] But both are amazing.

[50] Anyway, there are actually nine other countries with their own version of the office.

[51] You can get access to them with no geo -restriction when you use ExpressVPN.

[52] It lets you control where you want sites to think you're located.

[53] You can choose from nearly 100 different countries, giving you access to content that isn't available in your region.

[54] So again, get it on any device at ExpressVPN .com slash LexPod to get an extra three months free end to support this podcast.

[55] And now here's my conversation with Russ.

[56] Tedrick.

[57] What is the most beautiful motion of an animal or robot that you've ever seen?

[58] I think the most beautiful motion of a robot has to be the passive dynamic walkers.

[59] I think there's just something fundamentally beautiful.

[60] The ones in particular that Steve Collins built with Andy Rowena at Cornell, a 3D walking machine, so it was not confined to a boom or a plane, that you put it on top of a small ramp, give it a little push.

[61] It's powered only by gravity.

[62] No controllers, no batteries whatsoever.

[63] It just falls down the ramp.

[64] And at the time, it looked more natural, more graceful, more human -like than any robot we'd seen to date, powered only by gravity.

[65] How does it work?

[66] Well, okay, the simplest model, it's kind of like a slinky.

[67] It's like an elaborate slinky.

[68] One of the simplest models we used to think about it is actually a rimless wheel.

[69] So imagine taking a bicycle wheel, but take the rim off.

[70] So it's now just got a bunch of spokes.

[71] If you give that a push, it still wants to roll down the ramp.

[72] But every time its foot, its spoke comes around and hits the ground, it loses a little energy.

[73] Every time it takes a step forward, it gains a little energy.

[74] Those things can come into perfect balance, and actually they want to.

[75] It's a stable phenomenon.

[76] If it's going too slow, it'll speed up.

[77] If it's going too fast, it'll slow down, and it comes into a stable periodic motion.

[78] Now you can take that rimless wheel, which doesn't look very much like a human walking, take all the extra spokes away, put a hinge in the middle.

[79] Now it's two legs.

[80] That's called our compass gate walker.

[81] That can still, you give it a little push, starts falling down a ramp.

[82] Looks a little bit more like walking.

[83] At least it's a biped.

[84] But what Steve and Andy and Ted McGere started the whole exercise, But what Steve and Andy did was they took it to this beautiful conclusion, where they built something that had knees, arms, a torso, the arms swung naturally, give it a little push, and that looked like a stroll through the park.

[85] How do you design something like that?

[86] I mean, is that art or science?

[87] It's on the boundary.

[88] I think there's a science to getting close to the solution.

[89] I think there's certainly art in the way that they made a beautiful robot.

[90] but but then the finesse because because this was working they were working with a system that wasn't perfectly modeled wasn't perfectly controlled there's all these little tricks that you have to tune the suction cups at the knees for instance so that they stick but then they release at just the right time or there's all these little tricks of the trade which really are art but it was a point i mean it made the point we were at that time the walking robot the best walking robot in the world was Honda's Asimo.

[91] Absolutely Marvel of modern engineering.

[92] Is this 90s?

[93] This was in 97 when they first released, it sort of announced P2, and then it went through.

[94] It was Asimo by then in 2004.

[95] And it looks like this very cautious walking, like you're walking on hot coals or something like that.

[96] I think it gets a bad rap.

[97] Asimo is a beautiful machine.

[98] It does walk with its knees bent.

[99] Our Atlas walking had its knees bent.

[100] But actually, asthma was pretty fantastic.

[101] But it wasn't energy efficient.

[102] Neither was Atlas when we worked on Atlas.

[103] None of our robots that have been that complicated have been very energy efficient.

[104] But there's a thing that happens when you do control, when you try to control a system of that complexity.

[105] You try to use your motors to basically counteract gravity.

[106] Take whatever the world's doing to you and push back.

[107] erase the dynamics of the world and impose the dynamics you want because you can make them simple and analyzable mathematically simple and this was a very sort of beautiful example that you don't have to do that you can just let go let physics do most of the work right and you just have to give it a little bit of energy this one only walked down a ramp it would never walk on the flat to walk on the flat you have to give a little energy at some point but maybe instead of trying to take the forces imparted to you by the world and replacing them, what we should be doing is letting the world push us around, and we go with the flow.

[108] Very Zen, very Zen robot.

[109] Yeah, but, okay, so that sounds very Zen, but you can, I can also imagine how many, like, failed versions they had to go through.

[110] Like, how many, like, I would say it's probably, would you say it's in the thousands that they've had to have the system fall down before they figured out.

[111] I don't know if it's thousands, but it's a lot.

[112] It takes some patience.

[113] There's no question.

[114] So in that sense, control might help a little bit.

[115] Oh, the absolute, I think everybody, even at the time, said that the answer is to do with that with control.

[116] But it was just pointing out that maybe the way we're doing control right now isn't the way we should.

[117] Got it.

[118] So what about on the animal side, the ones that figured out how to move efficiently?

[119] Is there anything you find inspiring or beautiful?

[120] in the movement of any I do have a favorite example Okay So it sort of goes with the passive walking idea So is there You know how energy efficient are animals Okay there's a great series of experiments By George Lauder at Harvard And Mike Tranofilo at MIT They were studying fish swimming in a water tunnel Okay And one of these The type of fish they were studying Were these rainbow trout because there was a phenomenon well understood that rainbow trout when they're swimming upstream at mating season they kind of hang out behind the rocks and it looks like I mean that's tiring work swimming upstream they're hanging out behind the rocks maybe there's something energetically interesting there so they tried to recreate that they put in this water tunnel a rock basically a cylinder that had the same sort of vortex street the eddies coming off the back of the rock that you would see in a stream and they put a real fish behind this and watched how it swims.

[121] And the amazing thing is that if you watch from above, what the fish swims when it's not behind a rock, it has a particular gate.

[122] You can identify the fish the same way you look at a human walking down the street.

[123] You sort of have a sense of how a human walks.

[124] The fish has a characteristic gate.

[125] You put that fish behind the rock, its gate changes.

[126] And what they saw was that it was actually resonating and kind of surfing between the vortices.

[127] now here was the experiment that really was the clincher because there was still it wasn't clear how much of that was mechanics of the fish how much of that is control the brain so the clincher experiment and maybe one of my favorites to date although there are many good experiments they took this was now a dead fish they took a dead fish they put a string that went that tied the mouse of the fish to the rock so it couldn't go back and get caught in the grates.

[128] And then they asked, what would that dead fish do when it was hanging up behind the rock?

[129] And so what you'd expect is sort of flopped around like a dead fish in the vortex wake until something sort of amazing happens.

[130] And this video is worth putting in.

[131] Yeah.

[132] Right?

[133] What happens?

[134] The dead fish basically starts swimming upstream, right?

[135] It's completely dead.

[136] No brain, no motors, no control.

[137] but it's somehow the mechanics of the fish resonate with the vortex street and it starts swimming upstream it's one of the best examples ever what who do you get credit for that too is that just evolution constantly just figuring out by killing a lot of generations of animals like the most efficient motion is that uh or maybe the physics of our world completely like is like evolution applied not only to animals but just the entirety of it somehow drives to efficiency, like nature likes efficiency?

[138] I don't know if that question even makes any sense.

[139] I understand the question.

[140] That's reason.

[141] I mean, do they co -evolve?

[142] Yeah, somehow co -evol, yeah, like, I don't know if an environment can evolve, but, um, I mean, there are experiments that people do, careful experiments that show that, um, animals can adapt to unusual situations and recover efficiency.

[143] So there seems like at least in one direction, I, I think there is reason to believe that the animal's motor system and probably its mechanics adapt in order to be more efficient, but efficiency isn't the only goal, of course.

[144] Sometimes it's too easy to think about only efficiency, but we have to do a lot of other things first, not get eaten, and then all other things being equal, try to save energy.

[145] By the way, let's draw a distinction between control and mechanics.

[146] Like how would you define each?

[147] Yeah.

[148] I mean, I think part of the point is that we shouldn't draw a line as clearly as we tend to.

[149] But on a robot, we have motors and we have the links of the robot, let's say.

[150] If the motors are turned off, the robot has some passive dynamics.

[151] Okay.

[152] Gravity does the work.

[153] You can put springs.

[154] I would call that mechanics, right?

[155] If we have springs and dampers, which our muscles are springs and dampers and tendons.

[156] But then you have something that's doing active work.

[157] putting energy in, your motors on the robot, the controller's job is to send commands to the motor that add new energy into the system, right?

[158] So the mechanics and control interplay, somewhere the divide is around, you know, did you decide to send some commands to your motor, or did you just leave the motors off, let them do their work?

[159] Would you say, is most of nature on the dynamic side or the control side.

[160] So, like, if you look at biological systems, or, you know, we're living in a pandemic now, like, do you think a virus is a dynamic system or is there a lot of control, intelligence?

[161] I think it's both, but I think we maybe have underestimated how important the dynamics are, right?

[162] I mean, even our bodies, the mechanics of our bodies, certainly with exercise, they evolved.

[163] but so I actually I lost a finger in early 2000s and it's my fifth metacarpal and it turns out you use that a lot in ways you don't expect when you're opening jars even when I'm just walking around if I bump it on something there's a bone there that was used to taking contact my fourth metacarpal wasn't used to taking contact it used to hurt still does a little bit but actually my bone has remodeled, right?

[164] Over the last, over a couple of years, the geometry, the mechanics of that bone changed to address the new circumstances.

[165] So the idea that somehow it's only our brain that's adapting or evolving is, is not right.

[166] Maybe sticking on evolution for a bit, because it's tended to create some interesting things.

[167] Bipedal walking, do you, why the heck did evolution give?

[168] of us.

[169] I think we're, are we the only mammals that walk on two feet?

[170] No. I mean, there's a bunch of animals that do it a bit.

[171] I think we are the most successful bypass.

[172] I think some, uh, I think I read somewhere that, um, the reason the, you know, evolution made us walk on two feet is because there's an advantage to being able to carry food back to the tribe or something like that.

[173] So like you can carry, it's kind of this communal cooperative thing, so like to carry stuff back to a place of shelter and so on to share with others.

[174] Do you understand at all the value of walking on two feet from both the robotics and a human perspective?

[175] Yeah.

[176] There are some great books written about walking evolution of the human body.

[177] I think it's easy, though, to make bad evolutionary arguments.

[178] Sure.

[179] Most of them are probably bad, but what else can we do?

[180] I mean, I think a lot of what dominated our evolution probably was not the things that worked well sort of in the steady state, you know, when things are good.

[181] But for instance, people talk about what we should eat now because our ancestors were meat eaters or whatever.

[182] Oh, yeah, I love that.

[183] Yeah.

[184] But probably, you know, the reason that one pre -homosapian species versus another survived was not because of whether they ate well when there was lots of food.

[185] But when the Ice Age came, you know, probably one of them happened to be in the wrong place.

[186] One of them happened to forage a food that was okay even when the glaciers came or something like that.

[187] There's a million variables that contributed and we can't, and are actually the amount of information we're working with and telling these stories, these evolutionary stories is very little.

[188] So, yeah, just like you said, it seems like if we study history, it seems like history turns on like these little events that otherwise would seem meaningless, but in a grant, like when you, in retrospect, or turning points.

[189] absolutely and that's probably how like somebody got hit in the head with a rock because somebody slept with the wrong person back in the cave days and somebody get angry and that turned you know warring tribes combined with the environment all those millions of things and the meat eating which I get a lot of criticism because I don't know I don't know what your dietary processes are like but these days I've been eating only meat which is um there's There's a large community of people who say, yeah, probably make evolutionary arguments and say, you're doing a great job.

[190] There's probably an even larger community of people, including my mom, who says it's deeply unhealthy.

[191] It's wrong, but I just feel good doing it.

[192] But you're right.

[193] These evolutionary arguments can be flawed.

[194] But is there anything interesting to pull out for walking?

[195] There's a great book, by the way, a series of books by Nicholas Taylor, about Fooled by Randomness and Black Swan.

[196] I highly recommend them.

[197] But, yeah, they make the point nicely that probably it was a few random events that, yes, maybe it was someone getting hit by a rock, as you say.

[198] That said, do you think, I don't know how to ask this question or how to talk about this, but there's something elegant and beautiful about moving on two feet, obviously biased because I'm human.

[199] But from a robotics perspective, too, you work with robots on two feet.

[200] Is it all useful to build robots that are on two feet as opposed to four?

[201] Is there something useful about it?

[202] I think the most, I mean, the reason I spent a long time working on bipedal walking was because it was hard.

[203] And it challenged control theory in ways that I thought were important.

[204] I wouldn't have ever tried to convince you that you should start a company around bipeds or something like this.

[205] there are people that make pretty compelling arguments right i think the most compelling one is that the world is built for the human form and if you want a robot to work in the world we have today then you know having a human form is a pretty good way to go um and there are there are places that a biped can go that would be hard for uh other form factors to go even natural places but um you know at some point in the long run we'll be building our environments for our robots probably and so maybe that argument falls aside so you famously run barefoot oh do you still run barefoot I still run barefoot that's so awesome much to my wife's chagrin do you want to make an evolutionary argument for why running barefoot is advantageous what have you learned about human and robot movement in general from running barefoot human or robot and or Well, you know, it happened the other way, right?

[206] So I was studying walking robots, and I was, there's a great conference called the Dynamic Walking Conference, where it brings together both the biomechanics community and the walking robots community.

[207] And so I had been going to this for years and hearing talks by people who study barefoot running and other, the mechanics of running.

[208] So I did eventually read Born to Run.

[209] Most people read Born to Run.

[210] First, yeah.

[211] right um the other thing i had going for me is actually that i uh i wouldn't i wasn't a runner before and i learned to run after i had learned about barefoot running or i mean started running longer distances so i didn't have to unlearn and i'm definitely um i'm a big fan of it for me but i'm not going to i tend to not try to convince other people there's people who run beautifully with shoes on and that's good um but here's why it makes sense for me it's all about the long -term game, right?

[212] So I think it's just too easy to run 10 miles, feel pretty good, and then you get home at night and you realize my knees hurt.

[213] I did something wrong, right?

[214] If you take your shoes off, then if you hit hard with your foot at all, then it hurts.

[215] You don't like run 10 miles and then realize you've done some damage.

[216] you have immediate feedback telling you that you've done something that's that's maybe suboptimal and you change your gate I mean it's even subconscious if I right now having run many miles barefoot if I put a shoe on my gate changes in a way that I think is not as good so so it makes me land softer and I think my my goals for running are to do it for as long as I can into old age not to win any races and so for me this is a you know a way to protect myself yeah i think um first of all i've tried running barefoot many years ago uh probably the other way just just just uh reading born to run but just to understand because i felt like i couldn't put in the miles that i wanted to and it feels like um running for me and i think for a lot of people was one of those activities that we do often and we never really try to learn to do correctly.

[217] Like, it's funny, there's so many activities we do every day, like brushing our teeth, right?

[218] I think a lot of us, at least me, probably have never deeply studied how to properly brush my teeth, right?

[219] Or wash as now with a pandemic, or how to properly wash our hands.

[220] I do it every day, but we haven't really studied, like, am I doing this correctly?

[221] But running felt like one of those things that it was absurd not to study how to do correctly because it's the source of so much pain and suffering.

[222] Like I hate running, but I do it because I hate it, but I feel good afterwards.

[223] But I think it feels like you need to learn how to do it properly.

[224] So that's where barefoot running came in.

[225] And then I quickly realized that my gate was completely wrong.

[226] I was taking huge steps and landing hard on the heel, all those elements.

[227] And so, yeah, from that, I actually learned to take really small steps.

[228] Look, I already forgot the number, but I feel like it was 180 a minute or something like that.

[229] And I remember I actually just took songs that are 180 beats per minute and then tried to run at that beat.

[230] Just to teach myself, it took a long time.

[231] And I feel like after a while, you learn to run, you adjust it properly without going all.

[232] the way to barefoot, but I feel like barefoot is the legit way to do it.

[233] I mean, I think a lot of people would be really curious about it.

[234] If they're interested in trying, what would you, how would you recommend they start or try or explore?

[235] Slowly.

[236] That's the biggest thing people do is they are excellent runners and they're used to running long distances or running fast and they take their shoes off and they hurt themselves instantly trying to do something that they were used to doing.

[237] I think I lucked out in the sense that I couldn't run very far when I first started trying.

[238] And I run with minimal shoes, too.

[239] I mean, I will, you know, bring along a pair of actually like aqua socks or something like this.

[240] I can just slip on or running sandals.

[241] I've tried all of them.

[242] What's the difference between a minimal shoe and nothing at all?

[243] What's like feeling -wise?

[244] What does it feel like?

[245] There is a, I mean, I noticed my gate changing, right?

[246] So, I mean, your foot has as many molecules.

[247] and sensors as your hand does, right?

[248] Censors.

[249] Ooh, okay.

[250] And we do amazing things with our hands.

[251] And we stick our foot in a big solid shoe, right?

[252] So there's, I think, you know, when you're barefoot, you're just giving yourself more appropriate deception.

[253] And that's why you're more aware of some of the gate flaws and stuff like this.

[254] Now, you have less protection too.

[255] So, um, rocks and stuff.

[256] I mean, yeah, so, so I think people are who, are afraid of barefoot running or worry about getting cuts or getting stepping on rocks.

[257] First of all, even if that was a concern, I think those are all like a very short term.

[258] You know, if I get a scratch or something, it'll heal in a week.

[259] If I blow out my knees, I'm done running forever.

[260] So I will trade the short term for the long term anytime.

[261] But even then, you know, this, again, to my wife's chagrin, your feet get tough, right?

[262] A callous, okay.

[263] Yeah, I can run over almost anything now.

[264] I mean, what, maybe can you talk about, is there tips or tricks that you have suggestions about, like, if I wanted to try it?

[265] You know, there is a good book, actually.

[266] There's probably more good books since I read them.

[267] But Ken Bob, Barefoot Ken Bob, Saxton.

[268] He's an interesting guy.

[269] But I think his book captures.

[270] the right way to describe running, barefoot running to somebody, better than any other I've seen.

[271] So you run pretty good distances and you bike and is there, you know, if we talk about bucket list items, is there something crazy on your bucket list athletically that you hope to do one day?

[272] I mean, my commute is already a little crazy.

[273] What are we talking about here?

[274] What distance are we're talking about?

[275] Well, I live about 12 miles from MIT, but you can find lots of different ways to get there.

[276] So, I mean, I've run there for a long, many years, a bike there.

[277] All the ways?

[278] Yeah, but normally I would try to run in and then bike home, bike in, run home.

[279] But you have run there and back before?

[280] Sure.

[281] Barefoot?

[282] Yeah, or with minimal shoes or whatever that.

[283] 12 times two?

[284] Yeah.

[285] Okay.

[286] It became kind of a game of how can I get to work.

[287] I've rollerbladed.

[288] I've done all of weird stuff.

[289] But my favorite one these days I've been taking the Charles River to work.

[290] So I can put in a little rowboat, not so far from my house, but the Charles River takes a long way to get the MIT.

[291] So I can spend a long time getting there.

[292] And it's, you know, it's not about, I don't know, it's just about, I've had people ask me, how can you justify taking that time?

[293] but for me it's just a magical time to think to compress decompress um you know especially i'll wake up do a lot of work in the morning and then i kind of have to just let that settle before i'm ready for all my meetings and then on the way home it's a great time to let it sort of let that settle you you lead a like a large group of people i mean you're is there days where you're like, oh shit, I got to get to work in an hour.

[294] Like, I mean, is there, is there attention there where, and like if we look at the grand scheme of things, just like you said, long term, that meeting probably doesn't matter.

[295] Like, you can always say, I'll just, I'll run and let the meeting happen, how it happens.

[296] Like, what, how do you, that Zen, how do you, what do you do, what do you do, what do you do with tension between the real world saying urgently you need to be there this is important everything is melting down how we're going to fix this robot there's this critical meeting and then there's this the zen beauty of just running the simplicity of it you along with nature what do you do with that I would say I'm not a fast runner particularly probably my fastest splits ever was when I had to get to daycare on time because they were going to charge me you know some some dollar per minute that I was late I've run some fast splits to daycare, but that those times are past now.

[297] I think work, you can find a work -life balance in that way.

[298] I think you just have to.

[299] I think I am better at work because I take time to think on the way in.

[300] So I plan my day around it, and I rarely feel that those are really at odds.

[301] so what the bucket list item if we're talking 12 times two or approaching a marathon what have you run an ultra marathon before do you do races is there what's uh to win not to uh i'm not going to like take a dingy across the atlantic or something if that's what you want but uh but if someone does and wants to write a book i would totally read it because I'm a sucker for that kind of thing.

[302] No, I do have some fun things that I will try.

[303] I like to, when I travel, I almost always bike to Logan Airport and fold up a little folding bike and then take it with me and bike to wherever I'm going.

[304] And it's taken me, or I'll take a stand -up paddle board these days on the airplane and then I'll try to paddle around where I'm going or whatever.

[305] And I've done some crazy things.

[306] But not for the, you know, I've, I now talk, I don't know if you know who David Goggins is by any chance.

[307] Not well, but yeah.

[308] But I talk to him now every day, so he's the person who made me do this stupid challenge.

[309] So he's insane, and he does things for the purpose in the best kind of way.

[310] He does things like for the explicit purpose of suffering.

[311] Like he picks the thing that, like, whatever he thinks he can do, he does more.

[312] So is that, do you have that thing in you?

[313] or you uh i think it's become the opposite it's uh so you're like that dynamical system that the walk or the efficient uh yeah it's uh leave no pain right uh you should end feeling better than you started okay but um it's mostly i think and covid has tested this because i've lost my commute i think i'm perfectly happy walking around uh around town with my wife and uh kids if they could get them to go and it's more about just getting outside and getting away from the keyboard for some time just to let things compress.

[314] Let's go into robotics a little bit.

[315] What to you is the most beautiful idea in robotics?

[316] Whether we're talking about control or whether we're talking about optimization and the math side of things or the engineering side of things or the philosophical side of things.

[317] I think I've been lucky to experience something that not, So many roboticists have experienced, which is to hang out with some really amazing control theorists.

[318] And the clarity of thought that some of the more mathematical control theory can bring to even very complex, messy -looking problems is really, it really had a big impact on me. and I had a day even just a couple weeks ago where I had spent the day on a Zoom robotics conference having great conversations with lots of people.

[319] It felt really good about the ideas that were flowing and the like.

[320] And then I had a late afternoon meeting with one of my favorite control theorists.

[321] And we went from these abstract discussions about maybes and what ifs.

[322] And what a great idea to these super precise statements about systems that aren't that much more simple or abstract than the ones I care about deeply.

[323] And the contrast of that is, I don't know, it really gets me. I think people underestimate maybe the power of clear thinking.

[324] And so, for instance, deep learning is amazing.

[325] I use it heavily in our work.

[326] I think it's changed the world, unquestionable.

[327] It makes it easy to get things to work without thinking as critically about it.

[328] So I think one of the challenges as an educator is to think about how do we make sure people get a taste of the more rigorous thinking that I think goes along with.

[329] with some different approaches.

[330] Yeah, so that's really interesting.

[331] So understanding like the fundamentals, the first principles of the problem where in this case is mechanics, like how a thing moves, how a thing behaves, like all the forces involved, like really getting a deep understanding of that.

[332] I mean, from physics, the first principle thing come from physics, and here it's literally physics.

[333] physics.

[334] And this applies, in deep learning, this applies to, not just, I mean, it applies so cleanly in robotics, but it also applies to just in any data set.

[335] I find this true, I mean, driving as well.

[336] There's a lot of folks that work on autonomous vehicles that don't study driving, like, deeply.

[337] might be coming a little bit from the psychology side, but I remember I spent a ridiculous number of hours at lunch, at this lawn chair, and I would sit somewhere in MIT's campus there's a few interesting intersections, and we just watch people cross.

[338] So we were studying pedestrian behavior, and I felt like, as we record a lot of video, and then there's the computer vision extracts their movement, how they move their head, and so on.

[339] But, like, every time, I felt like I didn't understand enough.

[340] I just, I felt like I wasn't understanding what, how are people signaling to each other?

[341] What are they thinking?

[342] How cognizant are they of their fear of death?

[343] Like, what are we, like, what's the game, what's the underlying game theory here?

[344] What are, what are the, the incentives?

[345] And then I finally found a live stream of an intersection that's like high deaf that I just, I would watch.

[346] I wouldn't have to sit out there.

[347] But that's interesting.

[348] So, like, I feel...

[349] But that's tough.

[350] That's a tough example because, I mean, the learning...

[351] Humans are involved.

[352] Not just because human, but I think the learning mantra is that basically the statistics of the data will tell me things I need to know, right?

[353] And, you know, for the example you gave of all the nuances of, you know, eye contact or hand gestures or whatever that are happening for these subtle interactions between pedestrians and traffic.

[354] Right.

[355] Maybe the data will tell that story.

[356] Maybe even one level more meta than what you're saying.

[357] For a particular problem, I think it might be the case that data should tell us the story.

[358] But I think there's a rigorous thinking that is just an essential skill for a mathematician or an engineer that I just don't want to lose it.

[359] There are certainly super rigorous, rigorous control, or sorry, machine learning people.

[360] I just think deep learning makes it so easy to do some things that our next generation are not immediately rewarded for going through some of the more rigorous approaches.

[361] And then I wonder where that takes us.

[362] Well, I'm actually optimistic about it.

[363] I just want to do my part to try to steer that rigorous thinking.

[364] So there's like two questions I want to ask.

[365] Do you have sort of a good example of rigorous thinking where it's easy to get lazy and not do the rigorous thinking?

[366] And the other question I have is like, do you have advice of how to practice rigorous thinking in all the computer science disciplines that we've mentioned?

[367] Yeah, I mean, there are times where problems that can be solved with well -known mature methods could also be solved with a deep learning approach.

[368] And there's an argument that you must use learning even for the parts we already think we know because if the human has touched it, then you've biased the system and you've suddenly put a bottleneck in there that is your own mental model.

[369] But something like converting a matrix, you know, I think we know how to do that pretty well, even if it's a pretty big matrix, and we understand that pretty well.

[370] And you could train a deep network to do it, but you shouldn't, probably.

[371] So in that sense, rigorous thinking is understanding the scope and the limitations of the methods that we have, like how to use the tools of mathematics properly.

[372] Yeah.

[373] I think, you know, taking the...

[374] a class on analysis is all I'm sort of arguing is to take a chance to stop and force yourself to think rigorously about even you know the rational numbers or something you know it doesn't have to be the end -all problem but that exercise of clear thinking I think goes a long way and I just want to make sure we we keep preaching it don't lose it yeah but do you think when you're doing like rigorous thinking or like maybe trying to write down equations or to sort of explicitly, like, formally describe a system, do you think we'd naturally simplify things too much?

[375] Is that a danger you run into?

[376] Like, in order to be able to understand something about the system mathematically, we make it too much of a toy example.

[377] But I think that's the good stuff, right?

[378] That's how you understand the fundamentals?

[379] I think so.

[380] I think maybe even that's a key to intelligence or something.

[381] But, I mean, okay, what if Newton and Gell -Lah, Leo had deep learning.

[382] And they had done a bunch of experiments, and they told the world, here's your weights of your neural network.

[383] We've solved the problem.

[384] You know, where would we be today?

[385] I don't think we'd be as far as we are.

[386] There's something to be said about having the simplest explanation for a phenomenon.

[387] So I don't doubt that we can train neural networks to predict even physical, you know, F equals M .A. type of equations, but I maybe I want another Newton to come along because I think there's more to do in terms of coming up with the simple models for more complicated tasks.

[388] Yeah, let's not offend AI systems from 50 years now that are listening to this that are probably better at might be better at coming up with the F equals MA equations themselves.

[389] Oh, sorry, I actually think learning is probably a route to achieving this, but the representation matters, right?

[390] And I think having a function that takes my inputs to outputs that is arbitrarily complex may not be the end goal.

[391] I think there's still, you know, the most simple or parsimonious explanation for the data.

[392] Simple doesn't mean low -dimensional.

[393] That's one thing I think that we've, a lesson that we've learned.

[394] A standard way to do model reduction or system identification and controls is to the typical formulation is that you try to find the minimal state dimension realization of a system that hits some error bounds or something like that.

[395] And that's maybe not, I think we're learning that that state dimension is not the right metric.

[396] Of complexity.

[397] Of complexity.

[398] But for me, I think a lot about contact, the mechanics of contact.

[399] A robot hand is picking up an object or something.

[400] And when I write down the equations of motion for that, they look incredibly complex, not because actually not so much because of the dynamics of the hand when it's moving, but it's just the interactions and when they turn on and off, right?

[401] So having a high dimensional, you know, but simple description of what's happening out here is fine, but when I actually start touching, if I write down a, a different dynamical system for every polygon on my robot hand and every polygon on the object, whether it's in contact or not with all the combinatorics that explodes there, then that's too complex.

[402] So I need to somehow summarize that with a more intuitive physics way of thinking.

[403] And yeah, I'm very optimistic that machine learning will get us there.

[404] First of all, I mean, I'll probably do it in the introduction, but you're one of the great robotics people at MIT.

[405] You're a professor at MIT.

[406] You teach a lot of amazing courses.

[407] You run a large group.

[408] And you have an important history for MIT, I think, as being a part of the DARPA Robotics Challenge.

[409] Can you maybe first say, what is the DARPA Robotics Challenge and then tell your story around it, your journey with it?

[410] Yeah, sure.

[411] So the DARPA Robotics Challenge, It came on the tails of the DARPA Grand Challenge and DARPA Urban Challenge, which were the challenges that brought us, put a spotlight on self -driving cars.

[412] Gil Pratt was at DARPA and pitched a new challenge that involved disaster response.

[413] It didn't explicitly require humanoids, although humanoids came into the picture.

[414] This happened shortly after the Fukushima disaster in Japan, and our challenge was motivated roughly by that, because that was a case where if we had had robots that were ready to be sent in, there's a chance that we could have averted disaster, and certainly after the, in the disaster response, there were times we would have loved to have sent robots in.

[415] So in practice, what we ended up with was a grand challenge, a DARPA Robotics Challenge, where Boston Dynamics was to make humanoid robots.

[416] People like me and the amazing team at MIT were competing first in a simulation challenge to try to be one of the ones that wins the right to work on one of the Boston Dynamics humanoids in order to compete in the final challenge which was a physical challenge and at that point it was already so it was the side of this humanoid robots early on there were there were two tracks you could enter as a hardware team where you brought your own robot or you could enter through the virtual robotics challenge as a software team that would try to win the right to use one of the Boston Dynamics robots which are called Atlas Atlas humanoid robots yeah it was a 400 pound marvel but a you know pretty big, scary -looking robot.

[417] Expensive, too.

[418] Expensive at the time, yeah.

[419] Okay, so, I mean, how did you feel the prospect of this kind of challenge?

[420] I mean, it seems, you know, autonomous vehicles, yeah, I guess that sounds hard, but not really, from a robotics perspective.

[421] It's like, didn't they do it in the 80s?

[422] Is it kind of feeling I would have, like when you first look at the problem, and it's on wheels, but like, human or robots, that sounds really hard.

[423] So what, like, what are your, the, psychologically speaking, what were you feeling, excited, scared, why the heck did you get yourself involved in this kind of messy challenge?

[424] We didn't really know for sure what we were signing up for in the sense that you could have something that, as it was described in the call for participation, They could have put a huge emphasis on the dynamics of walking and not falling down and walking over rough terrain or the same description because the robot had to go into this disaster area and turn valves and pick up a drill, it cut the hole through a wall.

[425] It had to do some interesting things.

[426] The challenge could have really highlighted perception and autonomous planning or it ended up that, you know, locomoting over a complex terrain.

[427] played a pretty big role in the competition.

[428] And the degree of autonomy wasn't clear?

[429] The degree of autonomy was always a central part of the discussion.

[430] So what wasn't clear was how we would be able to how far we'd be able to get with it.

[431] So the idea was always that you want semi -autonomy, that you want the robot to have enough compute that you can have a degraded network link to a human.

[432] And so the same way we had degraded networks.

[433] at many natural disasters, you'd send your robot in, you'd be able to get a few bits back and forth, but you don't get to have enough financially to fully operate the robot, every joint of the robot.

[434] So, and then the question was, and the gamesmanship of the organizers, was to figure out what we're capable of, push us as far as we could, so that it would differentiate the teams that put more autonomy on the robot and had a few clicks and just said, go there, do this, there do this versus someone who's picking every footstep or something like that.

[435] So what were some memories, painful, triumphant from the experience?

[436] Like what was that journey?

[437] Maybe if you can dig in a little deeper, maybe even on the technical side, on the team side, that whole process of from the early idea stages to actually competing.

[438] I mean, this was a defining experience for me. It came at the right time for me in my career.

[439] I had gotten tenure before I was due a sabbatical, and most people do something relaxing and restorative for a sabbatical.

[440] So you got tenure before this?

[441] Yeah, yeah, yeah.

[442] It was a good time for me. We had a bunch of algorithms that we were very happy with.

[443] We wanted to see how far we could push them, and this was a chance to really test our metal, to do more proper software engineering.

[444] The team, we all just worked, our butts off.

[445] We, you know, we're in that lab almost all the time.

[446] Okay.

[447] So, I mean, there were some, of course, high highs and low lows throughout that.

[448] Anytime you're, you know, not sleeping and devoting your life to a 400 -pound humanoid.

[449] I remember actually one funny moment where we're all super tired.

[450] And so Atlas had to walk across cinder blocks.

[451] That was one of the obstacles.

[452] And I remember Atlas was powered down and hanging limp, you know, on the, on its harness and the humans were there like laying you know picking up and laying the brick down so that the robot could walk over it and I thought what is wrong with this you know we've got a robot just watching us do all the manual labor so that it can take its little um stroll across the terrain but um I mean even the even the virtual robotics challenge was was super nerve -wracking and dramatic I remember um so so we were using gazebo as a simulator on the cloud.

[453] There was all these interesting challenges.

[454] I think the investment that OSRF, C, whatever they were called at that time, Brian Gerke's team at open source robotics, they were pushing on the capabilities of gazebo in order to scale it to the complexity of these challenges.

[455] So, you know, up to the virtual competition.

[456] So the virtual competition was you will sign on at a certain time and we'll have a network connection to another machine on the cloud that is running the simulator of your robot.

[457] And your controller will run on this computer, this computer, and the physics will run on the other, and you have to connect.

[458] Now, the physics, they wanted it to run at real -time rates because there was an element of human interaction, and humans, if you do want to teleop, it works way better if it's at frame rate.

[459] Oh, cool.

[460] but it was very hard to simulate these complex scenes at real -time rate.

[461] So right up to like days before the competition, the simulator wasn't quite at real -time rate.

[462] And that was great for me because my controller was solving a pretty big optimization problem, and it wasn't quite at real -time rate.

[463] So I was fine.

[464] I was keeping up with the simulator.

[465] We were both running at about 0 .7.

[466] And I remember getting this email, and by the way, the perception folks, on our team hated that they knew that if my controller was too slow, the robot was going to fall down.

[467] And no matter how good their perception system was, if I can't make my controller fast.

[468] Anyways, we get this email like three days before the virtual competition.

[469] You know, it's for all the marbles.

[470] We're going to either get a humanoid robot or we're not.

[471] And we get an email saying, good news.

[472] We made the robot, does the simulator faster?

[473] It's now at one point.

[474] And I was just like, oh, man, what are we going to do here?

[475] So that came in late at night for me. A few days ahead.

[476] A few days ahead.

[477] I went over, it happened at Frank Permanter, who's a very, very sharp.

[478] He was a student at the time working on optimization.

[479] He was still in lab.

[480] Frank, we need to make the quadratic programming solver faster.

[481] Not like a little faster.

[482] It's actually, you know.

[483] And we wrote a new solver for that QP.

[484] together that night.

[485] And you solve her.

[486] So there's a really hard optimization problem that you're constantly solving.

[487] You didn't make the optimization problem simpler.

[488] You wrote a new solver.

[489] So, I mean, your observation is almost spot on.

[490] What we did was what everybody, I mean, people know how to do this, but we had not yet done this idea of warm starting.

[491] So we are solving a big optimization problem at every time step.

[492] But if you're running fast enough, the optimization.

[493] problem you're solving on the last time step is pretty similar to the optimization you're going to solve with the next.

[494] We, of course, had told our commercial solver to use warm starting, but even the interface to that commercial solver was causing us these delays.

[495] So what we did was we basically wrote, we called it fast QP at the time.

[496] We wrote a very lightweight, very fast layer, which would basically check if nearby solutions to the quadratic program, which were very very easily checked, could stabilize the robot.

[497] And if they couldn't, we would fall back to the solver.

[498] You couldn't really test this well, right?

[499] So we always knew that if we fell back to, if we, it got to the point where if for some reason things slowed down and we fell back to the original solver, the robot would actually literally fall down.

[500] So it was a harrowing sort of edge we were, ledge we were sort of on.

[501] But, I mean, actually, like the 400 -pound human.

[502] it could come crash into the ground if your solvers not fast enough.

[503] But, you know, we have lots of good experiences.

[504] So can I ask you a weird question?

[505] I get about the idea of hard work.

[506] So actually people, like students of yours that I've interacted with, and just in robotics, people in general, but they have moments, at moments of work, harder than most people I know in terms of if you look at different disciplines of how hard people work but they're also like the happiest like just like I don't know it's the same thing with like running people that push themselves to like the limit they all also seem to be like the most like full of life somehow and I get often criticized like you're not getting enough sleep what are you doing to your body blah blah blah like this kind of stuff and And I usually just kind of respond, like, I'm doing what I love.

[507] I'm passionate about.

[508] I love it.

[509] I feel like it's invigorating.

[510] I actually think, I don't think the lack of sleep is what hurts you.

[511] I think what hurts you is stress and lack of doing things that you're passionate about.

[512] But in this world, yeah, I mean, can you comment about why the heck robotics people are willing to push themselves to that degree.

[513] Is there value in that?

[514] And why are they so happy?

[515] I think you got it right.

[516] I mean, I think the causality is not that we work hard.

[517] And I think other disciplines work very hard too.

[518] But I don't think it's that we work hard and therefore we are happy.

[519] I think we found something that we're truly passionate about.

[520] It makes us very happy.

[521] And then we get a little involved with it and spend a lot of time on it.

[522] What a luxury to have something that you want to spend all your time on, right?

[523] We could talk about this for many hours, but maybe if we could pick, is there something on the technical side on the approach that you took that's interesting that turned out to be a terrible failure or a success that you carry into your work today about all the different ideas that were involved in making, whether in the simulation or in the real world making this semi -autonomous system work?

[524] I mean, it really did teach me something fundamental about what it's going to take to get robustness out of a system of this complexity.

[525] I would say the DARPA challenge really was foundational in my thinking.

[526] I think the autonomous driving community thinks about this.

[527] I think lots of people thinking about safety critical systems that might have machine learning in the loop are thinking about these questions.

[528] For me, the DARPA challenge was the moment where I realized, you know, we've spent every waking minute running this robot.

[529] And again, for the physical competition, days before the competition, we saw the robot fall down in a way it had never fallen down before.

[530] I thought, you know, how could we have found that?

[531] You know, we only have one robot.

[532] It's running almost all the time.

[533] We just didn't have enough hours in the day to test that robot.

[534] Something has to change, right?

[535] And I think that, I mean, I would say that the team that won was from Keist was the team that had two robots and was able to do not only incredible engineering, just absolutely top rate engineering, but also they were able to test at a rate and discipline that we didn't keep up with.

[536] What does testing look like?

[537] What are we talking about here?

[538] Like what's a loop of test?

[539] Like a, from start to finish, what is a loop of testing?

[540] Yeah, I mean, I think there's a whole philosophy to testing.

[541] There's the unit tests, and you can do that on a hardware.

[542] You can do that in a small piece of code.

[543] You write one function.

[544] You should write a test that checks that functions, input, and outputs.

[545] You should also write an integration test at the other extreme of running the whole system together, you know, where that try to turn on all the different functions that you think are correct.

[546] It's much harder to write the specifications for a system -level test, especially if that system is as complicated as a humanoid robot.

[547] But the philosophy is sort of the same.

[548] On the real robot, it's no different, but on a real robot, it's impossible to run the same experiment twice.

[549] So if you see a failure, you hope you caught something in the logs that tell you what happened, but you'd probably never be able to run exactly that experiment again.

[550] And right now, I think our philosophy is just.

[551] basically Monte Carlo estimation is just run as many experiments as we can, maybe try to set up the environment to make the things we are worried about happen as often as possible, but really we're relying on somewhat random search in order to test.

[552] Maybe that's all we'll ever be able to, but I think, you know, because there's an argument that the things that'll get you are the the things that are really nuanced in the world and there'd be very hard to for instance put back in a simulation yeah the i guess the edge cases what was the the hardest thing like so you said walking over rough terrain like the just taking footsteps i mean people there's it's so dramatic and painful in a certain kind of way to watch these videos from the DRC of robots falling yep i just so heartbreaking i don't know maybe it's because because, for me, at least, we anthropomorphize the robot.

[553] Of course, it's also funny for some reason.

[554] Like, humans falling is funny.

[555] For, I don't, it's some dark reason.

[556] I'm not sure why it is so, but it's also, like, tragic and painful.

[557] And so speaking of which, I mean, what made the robots fall and fail in your view?

[558] So I can tell you exactly what happened on a, I contributed one of those.

[559] Our team contributed one of those spectacular falls.

[560] Every one of those falls has a complicated story.

[561] I mean, at one time, the power effectively went out on the robot because it had been sitting at the door waiting for a green light to be able to proceed and its batteries, you know, and therefore it just fell backwards and smash its head to ground, and it was hilarious, but it wasn't because of bad software, right?

[562] But for ours, so the hardest part of the challenge, the hardest task, in my view, was getting out of the Polaris.

[563] It was actually relatively easy to drive the Polaris.

[564] Can you tell the story to interrupt the story of the car?

[565] People should watch this video.

[566] I mean, the thing you've come up with is just brilliant.

[567] But anyway, sorry, what's...

[568] Yeah, we kind of joke, we call it the big robot little car problem because somehow the race organizers decided to give us a 400 -pound humanoid.

[569] And they also provided the vehicle, which was a little Polaris.

[570] And the robot didn't really fit in the car.

[571] So you couldn't drive the car with your feet under the steering column.

[572] We actually had to straddle the main column of the, and have basically one foot in the passenger seat, one foot in the driver's seat, and then drive with our left hand.

[573] But the hard part was we had to then park the car, get out of the car.

[574] It didn't have a door.

[575] That was okay.

[576] But it's just getting up from crouched, from sitting.

[577] when you're in this very constrained environment.

[578] First of all, I remember after watching those videos, I was much more cognizant of how hard it is for me to get in and out of the car, and out of the car especially.

[579] Like, it's actually a really difficult control problem.

[580] Yeah.

[581] And I'm very cognizant of it when I'm, like, injured for whatever reason.

[582] Oh, that's really hard.

[583] Yeah.

[584] So how did you, how did you approach this problem?

[585] So we had a, you know, you think of NASA's operations and they have these checklists, you know, pre -launched checklist and the like, we weren't far off from that.

[586] We had this big checklist.

[587] And on the first day of the competition, we were running down our checklist.

[588] And one of the things we had to do, we had to turn off the controller, the piece of software that was running that would drive the left foot of the robot in order to accelerate on the gas.

[589] And then we turned on our balancing controller.

[590] And the nerves, jitters of the first day of the competition, someone forgot to check that box and turn that controller off.

[591] So we used a lot of motion planning to figure out a sort of configuration of the robot that we get up and over.

[592] We relied heavily on our balancing controller.

[593] And basically, when the robot was in one of its most precarious, you know, sort of configurations trying to sneak its big leg out of the side, the other controller that thought it was still driving told its left foot to go like this.

[594] And that wasn't good.

[595] But it turned disastrous for us because what happened was a little bit of push here.

[596] Actually, we have videos of us, you know, running into the robot with a 10 foot pole and it kind of will recover.

[597] But this is a case where there's no space to recover.

[598] So a lot of our secondary balancing mechanisms about like take a step to recover, they were all disabled because we were in the car.

[599] There's no place to step.

[600] So we were relying on our just lowest level reflexes.

[601] And even then, I think just hitting the foot on the floor, we probably could have recovered from it.

[602] But the thing that was bad that happened is when we did that and we jostled a little bit, the tailbone of our robot was only a little off the seat.

[603] It hit the seat.

[604] And the other foot came off the ground just a little bit.

[605] And nothing in our plans had ever told us what to do if your butts on the seat.

[606] and your feeder in the air.

[607] Feet in the air.

[608] And then the thing is, once you get off the script, things can go very wrong because even our state estimation, our system that was trying to collect all the data from the sensors and understand what's happening with the robot, it didn't know about this situation.

[609] So it was predicting things that were just wrong.

[610] And then we did a violent shake and fell off in our face first out of the robot.

[611] But like into the destination.

[612] That's true, we fell in, we got our point for egress.

[613] But so is there any hope for, that's interesting, is there any hope for Atlas to be able to do something when it's just on its butt and feet in the air?

[614] Absolutely.

[615] So you can, what do you?

[616] No, so that's, that is one of the big challenges.

[617] And I think it's still true, you know, Boston Dynamics and, and Amimal and there's this incredible work on, on legged robots happening around the world.

[618] most of them still are very good at the case where you're making contact with the world at your feet and they have typically point feet relatively their balls on their feet for instance if that if those robots get in a situation where the elbow hits the wall or something like this that's a pretty different situation now they have layers of mechanisms that will make I think the more mature solutions have have ways in which the controller won't do stupid things but a human for instance is able to leverage incidental contact in order to accomplish a goal.

[619] In fact, if you push me, I might actually put my handout and make a brand new contact.

[620] The feet of the robot are doing this on quadrupeds, but we mostly in robotics are afraid of contact on the rest of our body, which is crazy.

[621] There's this whole field of motion planning, collision -free motion planning, and we write very complex algorithms so that the robot can dance around and make sure it doesn't touch.

[622] the world.

[623] So people are just afraid of contact because contact the scene is a difficult.

[624] It's still a difficult control problem and sensing problem.

[625] Now you're a serious person.

[626] I'm a little bit of an idiot and I'm going to ask you some dumb questions.

[627] So I do martial arts.

[628] So like jihitsu, I wrestled my whole life.

[629] So let me let me ask the question.

[630] question, you know, like whenever people learn that I do any kind of AI or like I mentioned robots and things like that, they say, when are we going to have robots that, you know, that can win in a wrestling match or in a fight against a human.

[631] So we just mentioned sitting on your butt if you're in the air, that's a common position jiu -jitsu when you're on the you're a down opponent.

[632] Like, how difficult do you think is the problem?

[633] And when will we have a robot that can defeat a human in a wrestling match?

[634] And we're talking about a lot, like, I don't know if you're familiar with wrestling, but essentially it's basically the art of contact.

[635] It's like, because you're picking contact points and then, using like leverage like to off balance to to trick people like you make them feel like you're doing one thing and then they they change their balance and then you switch what you're doing and then results in a throw or whatever so like it's basically the art of multiple contexts so awesome it's a nice description of it so there's also an opponent in there right so so very dynamic Right.

[636] If you are wrestling a human and are in a game theoretic situation with a human, that that's still hard.

[637] But just to speak to the, you know, quickly reasoning about contact part of it, for instance.

[638] Yeah, maybe even throwing the game theory out of it, almost like, yeah, almost like a non -dynamic opponent.

[639] Right.

[640] There's reasons to be optimistic, but I think our best understanding of those problems are still pretty hard.

[641] I have been increasingly focused on manipulation, partly where that's a case where the contact has to be much more rich.

[642] And there are some really impressive examples of deep learning policies, controllers, that can appear to do good things through contact.

[643] We've even got new examples of deep learning models of predicting what's going to have.

[644] happen to objects as they go through contact.

[645] But I think the challenge you just offered there still alludes us, right?

[646] The ability to make a decision based on those models quickly.

[647] You know, I have to think, though, it's hard for humans too when you get that complicated.

[648] I think probably you had maybe a slow motion version of where you learned the basic skills and you've probably gotten better at it and there's much more subtle.

[649] but it might still be hard to actually, you know, really on the fly, take a, you know, model of your humanoid and figure out how to plan the optimal sequence.

[650] That might be a problem.

[651] We never solve.

[652] Well, the, I mean, one of the most amazing things to me about the, we can talk about martial arts.

[653] We could also talk about dancing.

[654] It doesn't really matter.

[655] Too human.

[656] I think it's the most interesting study of contact.

[657] It's not even the dynamic element of it.

[658] It's the like when you get good at it, it's so effortless.

[659] Like I can just, I'm very cognizant of the entirety of the learning process being essentially like learning how to move my body in a way that I could throw very large weights around effortlessly.

[660] Like, and I can feel the learning.

[661] Like I'm a huge believer in drilling of techniques.

[662] And you can just like feel your, I don't, you're not feeling, you're feeling, sorry, you're learning it intellectually a little bit, but a lot of it is the body learning it somehow, like instinctually.

[663] And whatever that learning is, that's really, I'm not even sure if that's, um, equivalent to, like a deep learning, learning a controller.

[664] I think it's something more, it feels like there's a lot of distributed learning going on.

[665] Yeah, I think there's, hierarchy and composition probably in the systems that we don't capture very well yet.

[666] You have layers of control systems.

[667] You have reflexes at the bottom layer and you have a system that's capable of planning a vacation to some distant country which is probably you probably don't have a control or a policy for every possible destination you'll ever pick.

[668] But there's something magical in the in -between and how do you go from these low -level feedback loops to something that feels like a pretty complex set of outcomes.

[669] You know, my guess is, I think there's evidence that you can plan at some of these levels, right?

[670] So Josh Tenenbaum just showed his talk the other day.

[671] He's got a game he likes to talk about, I think he calls it the pick three game or something, where he puts a bunch of clutter down in front of a person.

[672] and he says, okay, pick three objects.

[673] And it might be a telephone or a shoe or a Kleenex box or whatever.

[674] And apparently you pick three items and then you pick, he says, okay, pick the first one up with your right hand, the second one up with your left hand.

[675] Now using those objects, now as tools, pick up the third object.

[676] So that's down at the level of physics and mechanics and contact mechanics that I think we do learning, or we do have policies for.

[677] we do control for almost feedback, but somehow we're able to still, I mean, I've never picked up a telephone with a shoe and a water bottle before and somehow, and it takes me a little longer to do that the first time, but most of the time we can sort of figure that out.

[678] So, yeah, I think the amazing thing is this ability to be flexible with our models, plan when we need to use our well -oiled controllers when we don't, when we're in familiar territory.

[679] And having models, I think the other thing you just said was something about, I think your awareness of what's happening is even changing as you get, as you improve your expertise, right?

[680] So maybe you have a very approximate model of the mechanics to begin with.

[681] And as you gain expertise, you get a more refined version of that model.

[682] You're aware of muscles or balanced components that you just weren't even aware of before.

[683] So how do you scaffold that?

[684] yeah plus the fear of injury the ambition of goals of excelling and fear of mortality let's see what else is in there as motivations uh overinflated ego in the beginning uh like and then the crash of confidence in the middle all of those seem to be essential for the learning process and also and if all that's good then you're probably optimizing energy efficiency.

[685] Yeah, right.

[686] So we have to get that right.

[687] So, you know, there was this idea that you would have robots play soccer better than human players by 2050.

[688] That was the goal.

[689] Well, basically, was the goal to beat world champion team to become a World Cup, beat like a World Cup level team?

[690] So are we going to see that first or a robot?

[691] If you're familiar, there's an organization called UFC for mixed martial arts.

[692] Are we going to see a World Cup championship soccer team that are robots or a UFC champion mixed martial artist?

[693] That's a robot.

[694] I mean, it's very hard to say one thing is harder.

[695] Some problems harder than the other.

[696] What probably matters is who started the organization that, I mean, I think RoboCup has a pretty serious following.

[697] And there is a history now of people playing that game, learning about that game, building robots to play that game, building increasingly more human robots, it's got momentum.

[698] So if you want to have mixed martial arts compete, you better start your organization now, right?

[699] I think almost independent of which problem is technically harder because they're both hard and they're both different.

[700] That's a good point.

[701] I mean, those videos are just hilarious, especially the humanoid robots.

[702] trying to play soccer.

[703] I mean, they're kind of terrible right now.

[704] I mean, I guess there is robo -sumo wrestling.

[705] There's like the Robo -One competitions where they do have these robots that go on a table and basically fight.

[706] So maybe I'm wrong.

[707] Maybe...

[708] First of all, do you have a year in mind for a Robo Cup?

[709] Just from a robotics perspective, it seems like a super exciting possibility that, like, in the physical space, This is what's interesting.

[710] I think the world is captivated.

[711] I think it's really exciting.

[712] It inspires just a huge number of people when a machine beats a human at a game that humans are really damn good at.

[713] So you're talking about chess and go, but that's in the world of digital.

[714] I don't think machines have beat humans at a game in the physical space yet, but that would be just you have to make the rules very carefully right i mean if if atlas kick me in the shins i'm down and uh you know and and game over so there's you know it's it's very subtle on yeah i think the fair i think the fighting one is a weird one yeah because uh you're talking about a machine that's much stronger than you but yeah in terms of soccer basketball all those kind of soccer right i mean as soon as there's contact or whatever and there's there are some things that the robot will do better.

[715] I think if you really set yourself up to try to see could robots win the game of soccer as the rules were written, the right thing for the robot to do is to play very differently than a human would play.

[716] You're not going to get, you know, the perfect soccer player robot.

[717] You're going to get something that exploits the rules, exploits its super actuators, it's super low bandwidth, you know, feedback loops or whatever, and it's going to play the game differently than you want it to play.

[718] And I bet there's ways, I bet there's loopholes, right?

[719] We saw that in the DARPA Challenge, that it's very hard to write a set of rules that someone can't find a way to exploit.

[720] Let me ask another ridiculous question.

[721] I think this might be the last ridiculous question, but I doubt it.

[722] I aspire to ask as many ridiculous questions of, of a brilliant MIT professor.

[723] Okay, I don't know if you've seen the black mirror.

[724] It's funny.

[725] I never watched the episode.

[726] I know when it happened, though, because I gave a talk to some MIT faculty one day on an unassuming Monday or whatever, I was telling him about the state of robotics.

[727] And I showed some video from Boston Dynamics of the Quadruped spot at the time.

[728] It was the early version of, spot and there was a look of horror that went across the room and I said what you know I've I've I've shown videos like this a lot of times what happened and it turns out that this video had got yeah this black mirror episode had changed the way people watched um yeah the videos I was putting out the way they see these kinds of robots so I talked to so many people who are just terrified because of that episode probably of these kinds of robots they I almost want to say they almost kind of like enjoy being terrified I don't even know what it is about human psychology that kind of imagine doomsday the destruction of the universe or our society and kind of like enjoy being afraid i don't want to simplify it but it feels like they talk about it so often it almost there does seem to be an addictive quality to it um i talked to a guy that's a guy named jo rogan who's kind of the flag bearer for being terrified at these robots.

[729] Do you have two questions.

[730] One, do you have an understanding of why people are afraid of robots?

[731] And the second question is, in Black Mirror, just to tell you the episode, I don't even remember it that much anymore, but these robots, I think they can shoot like a pellet or something.

[732] They basically have, it's basically a spot with a gun.

[733] And how far are we away from having robots that go rogue like that, you know, basically spot that goes rogue for some reason and somehow finds a gun.

[734] Right.

[735] So, I mean, I'm not a psychologist.

[736] I think I don't know exactly why people react the way they do.

[737] I think we have to be careful about the way robots influence our society.

[738] and the like.

[739] I think that's something that's a responsibility that roboticists need to embrace.

[740] I don't think robots are going to come after me with a kitchen knife or a pallet gun right away.

[741] And I mean, if they were programmed in such a way, but I used to joke with Atlas that all I had to do was run for five minutes and its battery would run out.

[742] But actually, they've got a very big battery in there by the end.

[743] So it was over an hour.

[744] I think the fear is a bit cultural, though.

[745] Because, I mean, you notice that, like, I think in my age, in the U .S., we grew up watching Terminator, right?

[746] If I had grown up at the same time in Japan, I probably would have been watching Astro Boy.

[747] And there's a very different reaction to robots in different countries, right?

[748] So I don't know if it's a human innate fear of metal marvels or if it's something that we've done to ourselves with our sci -fi.

[749] Yeah, the stories we tell ourselves through movies, through just popular media.

[750] But if I were to tell, you know, if you were my therapist and I said, I'm really terrified that we're going to have these robots very soon that will hurt us.

[751] Like, how do you approach making me feel better?

[752] Like, why shouldn't people be afraid?

[753] There's a, I think there's a video that went viral recently.

[754] Everything, everything was spot in Boston Vegas goes viral in general, but usually it's like really cool stuff, like they're doing flips and stuff or like sad stuff.

[755] Atlas being hit with a broomstick or something like that.

[756] But there's a video where I think one of the new productions bought robots, which are awesome, it was like patrolling somewhere in like in some country.

[757] and like people immediately were like saying like this is like the dystopian future like the surveillance state for some reason like you can just have a camera like something about spot being able to walk on four feet with like really terrified people so like what what do you say to those people i think there is a legitimate fear there because so much of our future is uncertain um but at the same same time, technically speaking, it seems like we're not there yet.

[758] So what do you say?

[759] I mean, I think technology is complicated.

[760] It can be used in many ways.

[761] I think there are purely software attacks that somebody could use to do great damage.

[762] Maybe they have already.

[763] You know, I think wheeled robots could be used in bad ways too.

[764] Drones, right?

[765] I don't think that, let's see, I don't want to be building technology just because I'm compelled to build technology and I don't think about it.

[766] But I would consider myself a technological optimist, I guess, in the sense that I think we should continue to create and evolve and our world will change.

[767] And if we will introduce new challenges, we'll screw something up maybe.

[768] But I think also we'll invent ourselves out of those challenges and life will go on.

[769] So it's interesting because you didn't mention like this is technically too hard.

[770] I don't think robots are, I think people attribute a robot that looks like an animal as maybe having a level of self -awareness or consciousness or something that they don't have yet.

[771] right?

[772] So it's not, I think our ability to anthropomorphize those robots is probably we're assuming that they have a level of intelligence that they don't yet have.

[773] And that might be part of the fear.

[774] So in that sense, it's too hard.

[775] But, you know, there are many scary things in the world, right?

[776] So I think we're right to ask those questions.

[777] We're Right.

[778] In the short term, as we're working on it for sure, is there something long -term that scares you about our future with AI and robots?

[779] A lot of folks, from Elon Musk to Sam Harris, to a lot of folks talk about the existential threats about artificial intelligence.

[780] oftentimes robots kind of inspire that the most because of the anthropomorphism.

[781] Do you have any fears?

[782] It's an important question.

[783] I actually, I think I like Rod Brooks' answer maybe the best on this.

[784] I think, and it's not the only answer he's given over the years, but maybe one of my favorites is he says it's not going to be, he's got a book Flesh and Machines, I believe.

[785] It's not going to be the robots versus the people.

[786] We're all going to be robot people because, you know, we already have smartphones.

[787] Some of us have serious technology implanted in our bodies already, whether we have a hearing aid or a pacemaker or anything like this.

[788] People with amputations might have prosthetics.

[789] That's a trend, I think, that is likely to continue.

[790] I mean, this is now wild speculation.

[791] But, I mean, when do we get to cognitive implants and the like?

[792] Yeah, with neuralink, brain computer interfaces.

[793] That's interesting.

[794] So there's a dance between humans and robots.

[795] It's going to be impossible to be scared of the other out there, the robot, because the robot will be part of us, essentially.

[796] you'd be so intricately sort of part of our society that it might not even be implanted part of us but just it's so much a part of our yeah our society so in that sense the smartphone is already the robot we should be afraid of yeah uh i mean yeah and that all the usual fears arise of the misinformation the manipulation all those kinds of things that um that um that The problems are all the same.

[797] They're human problems, essentially, it feels like.

[798] Yeah, I mean, I think the way we interact with each other online is changing the value we put on, you know, personal interaction.

[799] And that's a crazy big change that's going to happen and rip through our society, right?

[800] And that has implications that are massive.

[801] I don't know if they should be scared of it or go with the flow, but I don't see, you know, some battle lines between, humans and robots being the first thing to worry about.

[802] I mean, I do want to just, as a kind of comment, maybe you can comment about your just feelings about Boston Dynamics in general.

[803] But, you know, I love science.

[804] I love engineering.

[805] I think there's so many beautiful ideas in it.

[806] And when I look at Boston Dynamics or Legged Robots in general, I think they inspire people curiosity and feelings in general, excitement about engineering more than almost anything else in popular culture.

[807] And I think that's such an exciting responsibility and possibility for robotics.

[808] And Boston Dynamics is riding that wave pretty damn well.

[809] Like they found it, they've discovered that hunger and curiosity and the people and they're doing magic with it.

[810] I don't care if the, I mean, I guess they're their company they have to make money, right?

[811] but they're already doing incredible work and inspiring the world about technology.

[812] I mean, do you have thoughts about Boston Dynamics and maybe others, your own work and robotics and inspiring the world in that way?

[813] I completely agree.

[814] I think Boston Dynamics is absolutely awesome.

[815] I think I show my kids those videos, you know, and the best thing that happens is sometimes they've already seen them, you know, right?

[816] I just think it's a pinnacle of success in robotics that is just one of the best things that's happened.

[817] I absolutely completely agree.

[818] One of the heartbreaking things to me is how many robotics companies fail.

[819] How hard it is to make money with the robotics company.

[820] Like I -Robot went through hell just to arrive at a Roomba to figure out one product.

[821] And then there's so many home robotics companies like GBO and Anki, Anki, the cutest toy.

[822] That's a great robot, I thought, went down.

[823] I'm forgetting a bunch of them.

[824] But a bunch of robotics companies fail.

[825] Rod's company, Rethink Robotics.

[826] Like, do you have anything hopeful to say about the possibility of making money?

[827] robots?

[828] Oh, I think you can't just look at the failures.

[829] I mean, Boston Dynamics is a success.

[830] There's lots of companies that are still doing amazingly good work in robotics.

[831] I mean, this is the capitalist ecology or something, right?

[832] I think you have many companies.

[833] You have many startups and they push each other forward and many of them fail and some of them get through and that's sort of the natural way of those things.

[834] I don't know that is robotic.

[835] really that much worse.

[836] I feel the pain that you feel too.

[837] Every time I read one of these, sometimes it's friends and I definitely wish it went better or went differently.

[838] But I think it's healthy and good to have bursts of ideas, bursts of activities.

[839] Ideas, if they are really aggressive, they should fail sometimes.

[840] Certainly, that's the research mantra, right?

[841] If you're succeeding at every problem you attempt, then you're not choosing aggressively enough.

[842] Is it exciting to you, the new spot?

[843] Oh, it's so good.

[844] When are you getting them as a pet or it?

[845] Yeah, I mean, I have to dig up 75K right now.

[846] I mean, it's so cool that there's a price tag, you can go and then actually buy it.

[847] I have a Skydeo R1.

[848] Love it.

[849] So, no, I would, I would absolutely be a customer.

[850] I wonder what your kids would think about I actually Zach from Boston Dynamics would let my kid drive in one of their demos one time and that was just so good so good I'll forever be grateful for that and there's something magical about the anthropomorphization of that arm it adds another level of human connection I'm not sure we understand from a control aspect, the value of anthropomorphization.

[851] I think that's an understudied and under understood engineering problem.

[852] It's been a psychologist have been studying it.

[853] I think it's part like manipulating our mind to believe things is a valuable engineering.

[854] Like this is another degree of freedom that can be controlled.

[855] I like that.

[856] Yeah.

[857] I think that's right.

[858] I think, you know, there's something that humans seem to do, or maybe my dangerous introspection, is I think we are able to make very simple models that assume a lot about the world very quickly.

[859] And then it takes us a lot more time like you're wrestling.

[860] You know, you probably thought you knew what you're doing with wrestling and you were fairly functional as a complete wrestler.

[861] And then you slowly got more expertise.

[862] So maybe it's natural that our first level of defense against seeing a new robot is to think of it in our existing models of how humans and animals behave.

[863] And as you spend more time with it, then you'll develop more sophisticated models that will appreciate the differences.

[864] Exactly.

[865] Can you say what does it take to control a robot?

[866] Like what is the control problem of a robot?

[867] and in general, what is a robot in your view?

[868] Like, how do you think of this system?

[869] What is a robot?

[870] What is a robot?

[871] I think robotics...

[872] I told the ridiculous questions.

[873] No, no, it's good.

[874] I mean, there's standard definitions of combining computation with some ability to do mechanical work.

[875] I think that gets us pretty close.

[876] But I think robotics has this problem that once things really work, we don't call them robots anymore like my dishwasher at home is pretty sophisticated beautiful mechanisms there's actually a pretty good computer probably a couple chips in there doing amazing things we don't think of that as a robot anymore which isn't fair because then roughly it means that robots robotics always has to solve the next problem and doesn't get to celebrate its past successes I mean even factory room floor robots are super successful They're amazing, but that's not the ones, I mean, people think of them as robots, but they don't, if you ask, what are the successes of robotics, somehow it doesn't come to your mind immediately.

[877] So the definition of robots, a system with some level automation that fails frequently.

[878] Something like, it's the computation plus mechanical work and an unsolved problem.

[879] That's all problem, yeah.

[880] So, so from our perspective of control and, um, mechanics, dynamics, what is a robot?

[881] So there are many different types of robots.

[882] The control that you need for a jebo robot, you know, some robot that's sitting on your countertop and interacting with you but not touching you, for instance, is very different than what you need for an autonomous car or an autonomous drone.

[883] It's very different than what you need for a robot that's going to walk or pick things up with its hands, right?

[884] But my passion has always been for the places where you're interacting more, you're doing more dynamic interactions with the world.

[885] So walking now manipulation.

[886] And the control problems there are beautiful.

[887] I think contact is one thing that differentiates them from many of the control problems we've solved classically, right?

[888] The modern control grew up stabilizing fighter jets that were passively unstable.

[889] and there's like amazing success stories from control all over the place.

[890] Power grid.

[891] I mean, there's all kinds of, it's everywhere that we don't even realize, just like AI is now.

[892] So you mentioned contact.

[893] Like, what's contact?

[894] So an airplane is an extremely complex system or a spacecraft landing or whatever.

[895] But at least it has the luxury of things change relatively continuously.

[896] That's an oversimplification.

[897] But if I make a small change in the command I send to my actuator, then the path that the robot will take tends to take a change only by a small amount.

[898] And there's a feedback mechanism here.

[899] And there's a feedback mechanism.

[900] And thinking about this as locally, like a linear system, for instance, I can use more linear algebra tools to study systems like that, generalizations of linear algebra to these smooth systems.

[901] What is contact?

[902] The robot has something very discontinuous that happens when it makes or breaks, when it starts touching the world.

[903] And even the way it touches or the order of contacts can change the outcome in potentially unpredictable ways.

[904] Not unpredictable, but complex ways.

[905] I do think there's a little bit of, a lot of people will say that contact is hard in robotics, even to simulate.

[906] And I think there's a little bit of a, there's truth to that, but maybe a misunderstanding around that.

[907] So what is limiting is that when we think about our robots and we write our simulators, we often make an assumption that objects are rigid.

[908] and when it comes down that their mass moves all it stays in a constant position relative to each other itself and that leads to some paradoxes when you go to try to talk about rigid body mechanics and contact and so for instance if I have a three -legged stool with just imagine it comes to a point at the at the legs so it's only touching the world at a point if I draw my high school physics diagram of this system then there's a couple of things that I'm given by by elementary physics I know if the system if the table is at rest if it's not moving zero velocities that means that the normal force all the forces are in balance so the force of gravity is being countered by the forces that the ground is pushing on my table legs I also know since it's not rotating that the moments have to balance.

[909] And since it can in, it's a three -dimensional table, it could fall in any direction.

[910] It actually tells me uniquely what those three normal forces have to be.

[911] If I have four legs on my table, four -legged table, and they were perfectly machined to be exactly the right, same height, and they're set down, and the table's not moving, then the basic conservation laws don't tell me there are many solutions for the forces that the ground could be putting on my legs that would still result in the table not moving.

[912] Now, the reason that seems fine, I could just pick one.

[913] But it gets funny now, because if you think about friction, what we think about with friction is our standard model says the amount of force that the table will push back, if I were to now try to push my table sideways, I guess I have a table here, is proportional to the normal force.

[914] So if I'm barely touching and I push, I'll slide.

[915] But if I'm pushing more and I push, I'll slide less.

[916] It's called Coolum Friction, is our standard model.

[917] Now, if you don't know what the normal force is on the four legs and you push the table, then you don't know what the friction forces are going to be.

[918] And so you can't actually tell the laws just aren't explicit yet about which way the table.

[919] is going to go.

[920] It could veer off to the left.

[921] It could veer off to the right.

[922] It could go straight.

[923] So the rigid body assumption of contact leaves us with some paradoxes, which are annoying for writing simulators and for writing controllers.

[924] We still do that sometimes because soft contact is potentially harder numerically or whatever, and the best simulators do both or do some combination of the two.

[925] But anyways, because of these kind of paradoxes, there's all kinds of paradoxes in contact, mostly due to these rigid body assumptions.

[926] It becomes very hard to, like, write the same kind of control laws that we've been able to be successful with for, like, fighter jets.

[927] We haven't been as successful writing those controllers for manipulation.

[928] And so you don't know what's going to happen at the point of contact, at the moment of contact.

[929] There are situations absolutely where you, where our laws, don't tell us.

[930] So the standard approach, that's okay.

[931] I mean, instead of having a differential equation, you end up with a differential inclusion, it's called.

[932] It's a set valued equation.

[933] It says that I'm in this configuration.

[934] I have these forces applied on me. And there's a set of things that could happen, right?

[935] And those aren't continuously.

[936] I mean, what, so when you're saying like non -smooth, they're not only not smooth, but this is discontinuous?

[937] The non -smooth comes in when I make or break a new contact first or when I transition from stick to slip.

[938] So you typically have static friction and then you'll start sliding and that'll be a discontinuous change in velocity, for instance, especially if you come to rest or...

[939] That's so fascinating.

[940] Okay, so what do you do?

[941] Sorry, I interrupted you.

[942] What's the hope under so much?

[943] uncertainty about what's going to happen?

[944] What are you supposed to do?

[945] I mean, control has an answer for this.

[946] Robust control is one approach, but roughly you can write controllers which try to still perform the right task despite all the things that could possibly happen.

[947] The world might want the table to go this way and this way, but if I write a controller that pushes a little bit more and pushes a little bit, I can certainly make the table go in the direction I want.

[948] It just puts a little bit more of a burden on the control system, right?

[949] And this discontinuities do change the control system because the way we write it down right now, every different control configuration, including sticking or sliding or parts of my body that are in contact or not, looks like a different system.

[950] And I think of them, I reason about them separately or differently, and the combinatorics of that blow up, right?

[951] So I just don't have enough time to compute all the possible contact configurations of my humanoid.

[952] Interestingly, I mean, I'm a humanoid.

[953] I have lots of degrees of freedom, lots of joints.

[954] I've only been around for a handful of years.

[955] It's getting up there.

[956] But I haven't had time in my life to visit all of the states in my system.

[957] Certainly all the contact configurations.

[958] So if step one is to consider every possible contact configuration that I'll ever be in, that's probably not a problem I need to solve.

[959] Just as a small tangent, what's a contact configuration?

[960] Just so we can enumerate, what are we talking about?

[961] How many are there?

[962] The simplest example maybe would be, imagine a robot with a flat foot.

[963] and we think about the phases of gate where the heel strikes and then the front toe strikes and then you can heel up, toe off.

[964] Those are each different contact configurations.

[965] I only had two different contacts, but I ended up with four different contact configurations.

[966] Now, of course, my robot might actually have bumps on it or other things, so it could be much more subtle than that, right?

[967] but it's just even with one sort of box interacting with the ground already in the plane has that many, right?

[968] And if I was just even a 3D foot, then probably my left toe might touch just before my right toe and things get subtle.

[969] Now, if I'm a dexterous hand and I go to talk about just grabbing a water bottle, if every, if I have to enumerate every possible order that my hand came into contact with the bottle, then I'm dead in the water any approach that we were able to get away with that in walking because we mostly touch the ground or then a small number of points, for instance, and we haven't been able to get dexterous hands that way.

[970] So, I mean, you've mentioned that people think that contact is really hard and that that's the reason that robotic manipulation is a problem is really hard.

[971] Is there any flaws in that thinking?

[972] So I think simulating contact is one aspect.

[973] And people often say that one of the reasons that we have a limit in robotics is because we do not simulate contact accurately in our simulators.

[974] And I think that is, the extent to which that's true is partly because our simulators, we haven't got mature enough simulators.

[975] there are some things that are still hard difficult that we should change but but we actually we know what the governing equations are they have some foibles like this indeterminacy but we should be able to simulate them accurately we have incredible open source community in robotics but it actually just takes a professional engineering team a lot of work to write a very good simulator like that now now where does I believe you've written, Drake?

[976] There's a team of people.

[977] I certainly spent a lot of hours on it myself.

[978] Well, what is Drake?

[979] What does it take to create a simulation environment for the kind of difficult control problems we're talking about?

[980] Right, so Drake is the simulator that I've been working on.

[981] There are other good simulators out there.

[982] I don't like to think of Drake as just a simulator.

[983] Because we write our controllers in Drake.

[984] We write our perception systems a little bit in Drake.

[985] But we write all of our, you know, low -level control and even planning and optimization.

[986] So it has optimization capabilities.

[987] Absolutely, yeah.

[988] I mean, Drake is three things, roughly.

[989] It's an optimization library, which is sits on, it provides a layer of abstraction in C++ and Python for commercial solvers.

[990] You can write linear programs, quadratic programs.

[991] you know, semi -definite programs, sums of squares programs, the ones we've used mixed integer programs, and it will do the work to curate those and send them to whatever the right solver is, for instance, and it provides a level of abstraction.

[992] The second thing is a system modeling language, a bit like LabView or Simulink, where you can make block diagrams out of complex systems, or it's like Ross in that sense, where you might have lots of Ross nodes that are each doing some part of your system.

[993] But to contrast it with Ross, we try to write, if you write a Drake system, then you have to, it asks you to describe a little bit more about the system.

[994] If you have any state, for instance, in the system, any variables that are going to persist, you have to declare them.

[995] Parameters can be declared and the like.

[996] But the advantage of doing that is that you can, if you like, run things all on one process, but you can also do control design against it.

[997] You can do, I mean, simple things like rewinding and playing back your simulations.

[998] For instance, you know, these things, you get some rewards for spending a little bit more upfront cost in describing each system.

[999] And I was inspired to do that because I think the complexity of Atlas, for instance, is just so great.

[1000] And I think although, I mean, Ross has been incredible, absolutely.

[1001] huge fan of what it's done for the robotics community, but the ability to rapidly put different pieces together and have a functioning thing is very good.

[1002] But I do think that it's hard to think clearly about a bag of disparate parts, Mr. Potatohead kind of software stack.

[1003] And if you can, you know, ask a little bit more out of each of those parts, then you can understand the way they work better.

[1004] You can try to verify them and the like.

[1005] You can do learning against them.

[1006] And then one of those systems, the last thing, I said the first two things that Drake is, but the last thing is that there is a set of multibody equations, rigid body equations, that is trying to provide a system that simulates physics.

[1007] And that, we also have renderers and other things, but I think the physics component of Drake is special in the sense that we have done excessive amount of engineering to make sure that we've written the equations correctly every possible tumbling satellite or spinning top or anything that we could possibly write as a test is tested we are making some you know I think fundamental improvements on the way you simulate contact just what does it take to simulate contact I mean it just seems I mean there's something just beautiful the way you were like explaining contact and you're like tapping your fingers on the on the table while you're doing it just um easily right easily just like just not even like it was like helping you think I guess what I see you have this like awesome demo of loading or unloading a dishwasher just picking up a plate grasping it like for the first time That just seems like so difficult.

[1008] How do you simulate any of that?

[1009] So it was really interesting that what happened was that we started getting more professional about our software development during the DARPA Robotics Challenge.

[1010] I learned the value of software engineering and how to bridle complexity.

[1011] I guess that's what I want to somehow fight against and bring some of the clear thinking of controls into these.

[1012] complex systems we're building for robots.

[1013] Shortly after the DARPA Robotics Challenge, Toyota opened a research institute, TRI, Toyota Research Institute.

[1014] They put one of their, there's three locations.

[1015] One of them is just down the street from MIT.

[1016] And I helped ramp that up right out as a part of my, the end of my sabbatical, I guess.

[1017] So TRI has given me the TRI robotics effort has made this investment in simulation in Drake and Michael Sherman leads a team there of just absolutely top -notch dynamics experts that are trying to write those simulators that can pick up the dishes.

[1018] And there's also a team working on manipulation there that is taking problems like loading the dishwasher and we're using that to study these really hard corner cases kind of problems in manipulation.

[1019] So for me, this, you know, simulating the dishes.

[1020] We could actually write a controller.

[1021] If we just cared about picking up dishes in the sink once, we could write a controller without any simulation whatsoever and we could call it done.

[1022] But we want to understand, like, what is the path you take to actually get to a robot that could perform that for any dish in anybody's kitchen with enough confidence that it could be a commercial product, right?

[1023] And it has deep learning perception in the loop.

[1024] It has complex dynamics in the loop.

[1025] It has controller.

[1026] It has a planner.

[1027] And how do you take all of that complexity and put it through this engineering discipline and verification and validation process to actually get enough confidence to deploy?

[1028] I mean, the, the, the, The DARPA challenge made me realize that that's not something you throw over the fence and hope that somebody will harden it for you, that there are really fundamental challenges in closing that last gap.

[1029] They're doing the validation and the testing.

[1030] I think it might even change the way we have to think about the way we write systems.

[1031] What happens if you have the robot running lots of tests, and it screws up.

[1032] it breaks a dish, right?

[1033] How do you capture that?

[1034] I said you can't run the same simulation or the same experiment twice in a real, on a real robot.

[1035] Do we have to be able to bring that one -off failure back into simulation in order to change our controllers, study it, make sure it won't happen again?

[1036] Do we, is it enough to just try to add that to our distribution and understand that on average we're going to cover that situation again?

[1037] there's like really subtle questions at the corner cases that I think we don't yet have satisfying answers for.

[1038] Like, how do you find the corner cases?

[1039] That's one kind of, is there, do you think that's possible to create a systematized way of discovering corner cases efficiently?

[1040] Yes.

[1041] In whatever the problem is.

[1042] Yes.

[1043] I mean, I think we have to get better at that.

[1044] I mean, control theory has for decades talked about.

[1045] about active experiment design.

[1046] What's that?

[1047] So people call it curiosity these days.

[1048] It's roughly this idea of trying to exploration or exploitation, but in the active experiment design is even more specific.

[1049] You could try to understand the uncertainty in your system, design the experiment that will provide the maximum information to reduce that uncertainty.

[1050] If there's a parameter you want to learn about, what is the optimal trajectory I could execute to learn about that parameter, for instance.

[1051] Scaling that up to something that has a deep network in the loop and a planning in the loop is tough.

[1052] We've done some work on, you know, with Matt O 'Kelly and Amoncina, we've worked on some falsification algorithms that are trying to do rare event simulation that try to just hammer on your simulator.

[1053] And if your simulator is good enough, you can, you can spend a lot of time, you know, You can write good algorithms that try to spend most of their time in the corner cases.

[1054] So you basically imagine you're building an autonomous car and you want to put it in, I don't know, downtown New Delhi all the time, right, an accelerated testing.

[1055] If you can write sampling strategies, which figure out where your controller is performing badly in simulation and start generating lots of examples around that.

[1056] you know it's just the space of possible places where that can be where things can go wrong is very big so it's hard to write those algorithms yeah rare rare event simulation is just like a really compelling notion uh if it's possible we we joked and we call it we call it the black swan generator it's black swan right because you don't just want the rare events you want the ones that are highly impactful i mean that's the most those are the most sort of profound questions we ask of world like what's the worst that can happen but what we're really asking isn't some kind of like computer science worst case analysis we're asking like what are the millions of ways this can go wrong and that's like our curiosity we humans I think are pretty bad at we just like run into it And I think there's a distributed sense because there's now like 7 .5 billion of us and so there's a lot of them and then a lot of them write blog posts about the stupid thing they've done so we learn in a distributed way.

[1057] There's some...

[1058] I think that's going to be important for robots too.

[1059] I mean, that's another massive theme at Toyota research for robotics is this fleet learning concept is, you know, the idea that I as a humanite don't have enough time to visit all of my states, right?

[1060] There's just a, it's very hard for one robot to experience all the things.

[1061] But that's not actually the problem we have to solve, right?

[1062] We're going to have fleets of robots that can have very similar appendages.

[1063] And at some point, maybe collectively they have enough data that their computational processes should be set up differently than ours, right?

[1064] It's a, have this vision of just, I mean, all these dish, washer unloading robots.

[1065] I mean, that robot dropping a plate and a human looking at the robot, probably pissed off.

[1066] Yeah.

[1067] But that's a special moment to record.

[1068] I think one thing in terms of fleet learning, and I've seen that because I've talked to a lot of folks, just like Tesla users or Tesla drivers, they're another company that's using this kind of fleet learning idea.

[1069] And one hopeful thing I have about humans is they really enjoy when a system improves learns.

[1070] So they enjoy fleet learning.

[1071] And the reason it's hopeful for me is they're willing to put up with something that's kind of dumb right now.

[1072] And they're like, if it's improving, they almost like enjoy being part of the, like, teaching it.

[1073] Almost like if you have kids, like you're teaching or something.

[1074] I think that's a beautiful thing because that gives me hope that we can put dumb robots out there.

[1075] I mean, the problem on the Tesla side with cars, cars can kill you.

[1076] That makes the problem so much harder.

[1077] Dishwasher unloading is a little safe.

[1078] That's why home robotics is really exciting.

[1079] And just to clarify, I mean, for people who might not know, I mean, TRI, Toyota Research Institute, So they're, I mean, they're pretty well known for like autonomous vehicle research, but they're also interested in home robotics?

[1080] Yep.

[1081] There's a big group working on, multiple groups working on home robotics.

[1082] It's a major part of the portfolio.

[1083] There's also a couple other projects and advanced materials discovery using AI and machine learning to discover new materials for car batteries and the like, for instance.

[1084] Yeah.

[1085] And that's been actually an incredibly successful team.

[1086] There's new projects starting up too.

[1087] Do you see a future of where robots are in our home and robots that have like actuators that look like arms in our home or like, you know, more like humanoid type robots?

[1088] Or is this, we're going to do the same thing that you just mentioned that, you know, the dishwasher is no longer a robot.

[1089] We're going to just not even see them as robots.

[1090] But, I mean, what's your vision of the home of the future, 10, 20 years from now, 50 years, if you get crazy?

[1091] Yeah, I think we already have Rumba's cruising around.

[1092] We have, you know, Alexis or Google Homes on our kitchen counter.

[1093] It's only a matter of time until they spring arms and start doing something useful like that.

[1094] So I do think it's coming.

[1095] I think lots of people have lots of motivations for doing it.

[1096] It's been super interesting actually learning about Toyota's vision for it, which is about helping people age in place.

[1097] Because I think that's not necessarily the first entry, the most lucrative entry point, but it's the problem maybe that we really need to solve no matter what.

[1098] So I think there's a real opportunity.

[1099] It's a delicate problem.

[1100] How do you work with people, help people, keep them active, engaged, but improve the quality of life and help them age in place, for instance.

[1101] It's interesting because older folks are also, I mean, there's a contrast there because they're not always the folks who are the most comfortable with technology, for example.

[1102] so there's a there's a division that's interesting there that you can do so much good with a robot for for uh older folks but there's a there's a gap to feel of understanding i mean it's actually kind of beautiful uh robot is learning about the human and the human is kind of learning about this new robot thing and it's uh also with um at least with uh like when i talk to my parents about robots.

[1103] There's a little bit of a blank slate there too.

[1104] Like, you can, I mean, they don't know anything about robotics.

[1105] So it's completely like wide open.

[1106] They don't have, they haven't, my parents haven't seen Black Mirror.

[1107] So like they, they, they, it's a blank slate.

[1108] Here's a cool thing.

[1109] Like, what can you do for me?

[1110] Yeah.

[1111] So it's an exciting space.

[1112] I think it's a really important space.

[1113] I do feel like, you know, a few years ago, uh, drones were successful enough in academia, they kind of broke out and started an industry and autonomous cars have been happening.

[1114] It does feel like manipulation.

[1115] In logistics, of course, first, but in the home shortly after, seems like one of the next big things that's going to really pop.

[1116] So I don't think we talked about it, but what's soft robotics?

[1117] So we talked about like rigid bodies like if we can just linger on this whole touch thing yeah so what's soft robotics so I told you that I really dislike the fact that robots are afraid of touching the world all over their body so there's a couple reasons for that if you look carefully at all the places that robots actually do touch the world they're almost always soft they have some sort of pad on their fingers or a rubber soul on their foot.

[1118] But if you look up and down the arm, we're just pure aluminum or something.

[1119] So that makes it hard, actually.

[1120] In fact, hitting the table with your, you know, your rigid arm or nearly rigid arm is a, has some of the problems that we talked about in terms of simulation.

[1121] I think it fundamentally changes the mechanics of contact when you're soft, right?

[1122] You turn point contacts into patch contacts, which can have torsional friction, you can have distributed load.

[1123] If I want to pick up an egg, right?

[1124] If I pick it up with two points, then in order to put enough force to sustain the weight of the egg, I might have to put a lot of force to break the egg.

[1125] If I envelop it with contact all around, then I can distribute my force across the shell of the egg and have a better chance of not breaking it.

[1126] So soft robotics is for me a lot about changing the mechanics of contact.

[1127] Does it make the problem a lot harder?

[1128] Quite the opposite.

[1129] It changes the computational problem.

[1130] I think because of the, I think our world and our mathematics has biased us towards rigid, but it really should make things better in some ways, right?

[1131] It's a, I think the future is unwritten there.

[1132] but the other thing ultimately you think ultimately you'll make things simpler if we embrace the softness of the world it makes things smoother right so the the result of small actions is less discontinuous but it also means potentially less you know instantaneously bad for instance I won't necessarily contact something and send it flying off The other aspect of it that just happens to dovetail really well is that soft robotics tends to be a place where we can embed a lot of sensors to.

[1133] So if you change your hardware and make it more soft, then you can potentially have a tactile sensor, which is measuring the deformation.

[1134] So there's a team at TRI that's working on soft hands, and you get so much more information.

[1135] if you can put a camera behind the skin roughly and get fantastic tactile information which is it's super important like in manipulation one of the things that really is frustrating is if you work super hard on your head mounted on your perception system for your head mounted cameras and then you've identified an object you reached down to touch it and the first the last thing that happens right before the most important time you stick your hand and you're occluding your head mounted sensors right So in all the part that really matters, all of your off -board sensors are, you know, are occluded.

[1136] And really, if you don't have tactile information, then you're blind in an important way.

[1137] So it happens that soft robotics and tactile sensing tend to go hand in hand.

[1138] I think we've kind of talked about it, but you taught a course on underactuated robotics.

[1139] I believe that was the name of it, actually.

[1140] That's right.

[1141] Can you talk about it in that context?

[1142] What is underactuated robotics?

[1143] Right.

[1144] So underactuated robotics is my graduate course.

[1145] It's online mostly now, in the sense that the lectures of it, I think.

[1146] Right.

[1147] It's really great.

[1148] I recommend it highly.

[1149] Look on YouTube for the 2020 versions until March and then you have to go back to 2019, thanks to COVID.

[1150] No, I've poured my heart into that.

[1151] class.

[1152] And lecture one is basically explaining what the word underactuated means.

[1153] So people are very kind to show up and then maybe have to learn what the title of the course means over the course of the first lecture.

[1154] That first lecture is really good.

[1155] You should watch it.

[1156] Thanks.

[1157] It's a strange name, but I thought it captured the essence of what control was good at doing and what control was bad at doing.

[1158] So what do I mean by under -actuated?

[1159] So a mechanical system has many degrees of freedom, for instance.

[1160] I think of a joint as a degree of freedom, and it has some number of actuators, motors.

[1161] So if you have a robot that's bolted to the table that has five degrees of freedom and five motors, then you have a fully actuated robot.

[1162] If you have, if you have, if you take away one of those motors, then you have an under -actuated robot.

[1163] Now, why on Earth, I have a good friend who likes to tease me, he said, Russ, if you had more research funding, would you work on fully actuated robots?

[1164] Yeah.

[1165] And the answer is no. The world gives us under -actuated robots, whether we like it or not.

[1166] I'm a human.

[1167] I'm an under -actuated robot, even though I have more muscles than my big degrees of freedom because I have in some places multiple muscles attached to the same joint.

[1168] But still, there's in a really important degree of freedom that I have, which is the location of my center of mass in space, for instance.

[1169] I can jump into the air and there's no motor that connects my center of mass to the ground in that case.

[1170] So I have to think about the implications of not having control over everything.

[1171] The passive dynamic walkers are the extreme view of that, where you've taken away all the motors and you have to let physics do the work.

[1172] But it shows up in all of the walking robots where you have to use some actuators to push and pull even the degrees of freedom that you don't have an actuator on.

[1173] That's referring to walking if you're like falling forward.

[1174] Like is there a way to walk that's fully actuated?

[1175] So it's a subtle point.

[1176] When you're in contact and you're, have your feet on the ground, there are still limits to what you can do, right?

[1177] Unless I have suction cups on my feet, I cannot accelerate my center of mass towards the ground faster than gravity because I can't get a force pushing me down, right?

[1178] But I can still do most of the things that I want to.

[1179] So you can get away with basically thinking of the system as fully actuated unless you suddenly needed to accelerate down super fast.

[1180] but as soon as I take a step I get into the more nuanced territory and to get to really dynamic robots or airplanes or other things I think you have to embrace the underactuated dynamics manipulation people think is manipulation under under actuated even if my arm is fully actuated I have a motor if my goal is to control the position and orientation of this cup that then I don't have an actuator for that directly.

[1181] So I have to use my actuators over here to control this thing.

[1182] Now it gets even worse.

[1183] Like, what if I have to button my shirt?

[1184] Okay.

[1185] What are the degrees of freedom of my shirt, right?

[1186] I suddenly, that's a hard question to think about.

[1187] It kind of makes me queasy as a, thinking about my states -based control ideas.

[1188] But actually, those are the problems that make me so excited about manipulation right now, is that it breaks a lot of the foundational control stuff that I've been thinking about.

[1189] Is there, what are some interesting insights you can say about trying to solve an underactuated control in an underactuated system?

[1190] So I think the philosophy there is let physics do more of the work.

[1191] The technical approach has been optimization.

[1192] So you typically formulate your decision making for, control as an optimization problem, and you use the language of optimal control, and sometimes numerous, often numerical optimal control, in order to make those decisions and balance, you know, these complicated equations of, and in order to control, you don't have to use optimal control to do underactuated systems, but that has been the technical approach that has borne the most fruit in our, at least in our line of work.

[1193] And there's some, so in underactuated systems, when you say, let physics do some of the work.

[1194] So there's a kind of feedback loop that observes the state that the physics brought you to.

[1195] So like you've, there's a perception there.

[1196] This is, there's a feedback somehow.

[1197] Do you ever loop in like complicated perception systems into this whole picture?

[1198] Right.

[1199] Right around the time of the DARPA challenge, we had a complicated perception system in the DARPA challenge.

[1200] We also started to embrace perception for our flying vehicles at the time.

[1201] We had a really good project on trying to make airplanes fly at high speeds through forests.

[1202] Sirtash Karaman was on that project, and it was a really fun team to work on.

[1203] He's carried it much farther forward since then.

[1204] And that's using cameras for perception?

[1205] So that was using cameras.

[1206] That was at the time we felt like LiDAR was too heavy and two power.

[1207] heavy to be carried on a light UAV and we were using cameras and that was a big part of it was just how do you do even stereo matching at a fast enough rate with a small camera, a small onboard compute.

[1208] Since then we have now, so the deep learning revolution unquestionably changed what we can do with perception for robotics and control.

[1209] So in manipulation we can address, we can use perception in, I think, a much deeper way.

[1210] And we get into not only, I think the first use of it, naturally, would be to ask your deep learning system to look at the cameras and produce the state, which is like the pose of my thing, for instance.

[1211] But I think we've quickly found out that that's not always the right thing to do.

[1212] Why is that?

[1213] Because what's the state of my shirt?

[1214] Imagine, I've...

[1215] It's very noisy, you mean?

[1216] If the first step of me trying to button my shirt is estimate the full state of my shirt, including, like, what's happening in the back here, whatever, whatever, that's just not the right specification.

[1217] There are aspects of the state that are very important to the task.

[1218] There are many that are unobservable and not important to the task.

[1219] So you really need, it begs new questions about state representation.

[1220] Another example that we've been playing with in lab has been just the idea of chopping onions, okay, or carrots, it turns out to be better.

[1221] So the onions stink up the lab, and they're hard to see in a camera.

[1222] The details matter, yeah.

[1223] Details matter, you know.

[1224] So if I'm moving around a particular object, right, then I think about, oh, it's got a position or an orientation in space, that's the description I want.

[1225] now when I'm chopping an onion okay the first chop comes down I have now a hundred pieces of onion does my control system really need to understand the position and orientation and even the shape of the hundred pieces of onion in order to make a decision probably not you know and if I keep going I'm just getting more and more is my state space getting bigger as I cut it's it it's not right so somehow there's a I think there's a richer idea of state.

[1226] It's not the state that is given to us by Lagrangian mechanics.

[1227] There is a proper Lagrangian state of the system, but the relevant state for this is some latent state is what we call it in machine learning, but there's some different state representation.

[1228] Some compressed representation.

[1229] And that's what I worry about saying compressed because because it doesn't, I don't mind that it's low dimensional or not, but it has to be something that's easier to think about.

[1230] By as humans.

[1231] Or my algorithms.

[1232] Or the algorithms being like control, optimal.

[1233] So for instance, if the contact mechanics of all of those onion pieces and all the permutations of possible touches between those onion pieces, you know, you can give me a high dimensional state representation.

[1234] I'm okay if it's linear.

[1235] But if I have to think about all the possible shattering combinatorics of that, then my robot's going to sit there thinking and the soup's going to get cold or something.

[1236] So since you taught the course, it kind of entered my mind, the idea of underactuated as really compelling to see the world in this kind of way.

[1237] Do you ever, you know, if we talk about onions or you talk about the world with people in it in general, Do you see the world as basically an under -actuated system?

[1238] Do you often look at the world in this way?

[1239] Or is this overreach?

[1240] Under -actuated as a way of life, man. Exactly.

[1241] I guess that's what I'm asking.

[1242] I do think it's everywhere.

[1243] I think in some places, we already have natural tools to deal with it.

[1244] You know, it rears its head.

[1245] I mean, in linear systems, it's not a problem.

[1246] We just, like an underactuated linear system is really not sufficiently distinct from a fully actuated linear system.

[1247] It's a subtle point about when that becomes a bottleneck and what we know how to do with control.

[1248] It happens to be a bottleneck, although we've gotten incredibly good solutions now, but for a long time that I felt that that was the key bottleneck in legged robots.

[1249] And roughly now, the underactuated course is, you know, me trying to tell people everything I can about.

[1250] how to make Atlas to a backflip, right?

[1251] I have a second course now that I teach in the other semesters, which is on manipulation, and that's where we get into now more of the, that's a newer class.

[1252] I'm hoping to put it online this fall completely.

[1253] And that's going to have much more aspects about these perception problems and the state representation questions and then how do you do control.

[1254] And the thing that's a little bit sad is that, for me at least, is there's a lot of manipulation tasks that people want to do and should want to do.

[1255] They could start a company with it and make very successful that don't actually require you to think that much about dynamics or dynamics at all even, but certainly underactuated dynamics.

[1256] Once I have, if I reach out and grab something, if I can sort of assume it's rigidly attached to my hand, then I can do a lot of interesting, meaningful things with it without really ever thinking about the dynamics of that object.

[1257] So we've built systems that kind of reduce the need for that, enveloping grasps and the like.

[1258] But I think the really good problems are manipulation.

[1259] So manipulation, by the way, is more than just pick and place.

[1260] That's like a lot of people think of that, just grasping.

[1261] I don't mean that.

[1262] I mean buttoning my shirt.

[1263] I mean tying shoelaces.

[1264] How do you program a robot to tie shoelaces?

[1265] And not just one shoe, but every shoe.

[1266] issue, right?

[1267] That's a really good problem.

[1268] It's tempting to write down like the infinite dimensional state of the of the laces.

[1269] That's probably not needed to write a good controller.

[1270] I know we could hand design a controller that would do it, but I don't want that.

[1271] I want to understand the principles that would allow me to solve another problem that's kind of like that.

[1272] But I think if we can stay pure in our approach, then the challenge of tying anybody's shoes is a great challenge that's a great challenge i mean and the soft touch comes into play there that's really interesting um let me ask another ridiculous question on this topic how important is touch we haven't talked much about humans but uh i have this argument with my dad where like i think you can fall in love with the robot based on uh language alone And he believes that touch is essential, a touch and smell, he says.

[1273] So in terms of robots, you know, connecting with humans, we can go philosophical in terms of like a deep meaningful connection, like love.

[1274] But even just like collaborating in an interesting way, how important is touch, like from an engineering perspective and a philosophical one?

[1275] I think it's super important.

[1276] Let's even just in a practical sense, if we forget about the emotional part of it, but for robots to interact safely while they're doing meaningful mechanical work in the, you know, close contact with or vicinity of people that need help, I think we have to have them, we have to build them differently.

[1277] They have to be afraid, not afraid of touching the world.

[1278] So I think Baymax is, just awesome.

[1279] That's just like the movie of Big Hero 6 and the concept of Baymax, that's just awesome.

[1280] I think we should, and we have some folks at Toyota that are trying to, Toyota research that are trying to build Baymax roughly.

[1281] And I think it's just a fantastically good project.

[1282] I think it will change the way people physically interact.

[1283] The same way, I mean, you gave a couple examples earlier.

[1284] But if the robot that was walking around my home looked more like a teddy bear and a little less like the Terminator, that could change completely the way people perceive it and interact with it.

[1285] And maybe they'll even want to teach it, like you said, right?

[1286] You could not quite gamify it, but somehow instead of people judging it and looking at it as if it's not doing as well as a human, they're going to try to help out the cute teddy bear, right?

[1287] Who knows?

[1288] But I think we're building robots wrong, and being more soft and more contact is important, right?

[1289] Yeah, like all the magical moments I can remember with robots, well, first of all, just visiting your lab and seeing Atlas, but also Spot Mini.

[1290] When I first saw Spot Mini in person and hung out with him, her, it, I don't have trouble in.

[1291] gendering robots.

[1292] I feel robotics people really say, always it.

[1293] I kind of like the idea that's a her or him.

[1294] There's a magical moment, but there's no touching.

[1295] I guess the question I have, have you ever been, like, have you had a human robot experience where a robot touched you?

[1296] And like, it was like, Wait, was there a moment that you've forgotten that a robot is a robot, and the anthropomorphization stepped in, and for a second you forgot that it's not human?

[1297] I mean, I think when you're in on the details, then we, of course, anthropomorphized our work with Atlas, but in, you know, in verbal communication and the like, I think we were pretty aware of it as a machine that.

[1298] needed to be respected.

[1299] I actually, I worry more about the smaller robots that could still, you know, move quickly if programmed wrong and we have to be careful actually about safety and the like right now.

[1300] And that, if we build our robots correctly, I think, then those, a lot of those concerns could go away.

[1301] And we're seeing that trend.

[1302] We're seeing the lower cost, lighter weight arms now that could be fundamentally safe.

[1303] I mean, I do think touch is so fundamental.

[1304] Ted Edelson is great.

[1305] He's a perceptual scientist at MIT, and he studied vision most of his life.

[1306] And he said, when I had kids, I expected to be fascinated by their perceptual development.

[1307] But what really, what he noticed felt more impressive, more dominant was the way that they would touch everything.

[1308] everything and pick things up and pick it on their tongue and whatever and he said watching his daughter convinced him that actually he needed to study tactile sensing more so there's something very important I think it's it's a little bit also of the passive versus active part of the world right you can passively perceive the world but it's fundamentally different if you can do an experiment, right?

[1309] And if you can change the world.

[1310] And you can learn a lot more than a passive observer.

[1311] So you can, in dialogue, that was your initial example, you could have an active experiment exchange.

[1312] But I think if you're just a camera watching YouTube, I think that's a very different problem than if you're a robot that can apply force and touch.

[1313] I think it's important.

[1314] Yeah, I think it's just an exciting area of research.

[1315] I think you're probably right that this hasn't been under -researched.

[1316] To me, as a person who's captivated by the idea of human -robot interaction, it feels like such a rich opportunity to explore touch.

[1317] Not even from a safety perspective, but like you said, the emotional, too.

[1318] I mean, safety comes first.

[1319] But the next step is like, you know, like a real human connection, even in the world, like even in the industrial setting.

[1320] It just feels like it's nice for the robot.

[1321] I don't know, you know, you might disagree with this, but because I think it's important to see robots as tools often, but I don't know.

[1322] I think they're just always going to be more effective once you humanize them.

[1323] Like, it's convenient now to think of them as tools because we want to focus on the safety.

[1324] But I think ultimately to create a good experience for the worker, for the person, there has to be a human element.

[1325] I don't know.

[1326] For me, it feels like an industrial robotic arm would be better if it has a human element.

[1327] I think, like, rethink robotics had that idea with Baxter and having eyes and so on, having, I don't know.

[1328] I'm a big believer in that.

[1329] It's not my area, but I am also a big believer.

[1330] Do you have an emotional connection to Alice?

[1331] Like, do you miss them?

[1332] I mean, yes, I don't know if I, more so than if I had a different science project that I worked on super hard, right?

[1333] But, yeah, I mean, the robot, we basically had to do heart surgery on the robot in the final competition, we melted the core and yeah there was something about watching that robot hanging there we know we had to compete with it in an hour and it was getting its cuts ripped out those are all historic moments i think if you look back like 100 years from now and yeah i think those are important moments in robotics i mean these are the early day you look at like the early days of a lot of scientific disciplines they look ridiculous there's full of failure but It feels like robotics will be important in the coming 100 years.

[1334] I think so.

[1335] And these are the early days.

[1336] So I think a lot of people look at a brilliant person such as yourself and are curious about the intellectual journey they've took.

[1337] Is there maybe three books, technical, fiction, philosophical, that had a big impact in your life that you would recommend perhaps other?

[1338] reading.

[1339] Yeah.

[1340] So I actually didn't read that much as a kid, but I read fairly voraciously now.

[1341] There are some recent books that if you're interested in this kind of topic, like AI Superpowers by Kifu Lee is just a fantastic read.

[1342] You must read that.

[1343] Uval Harari is just, I think that can open your mind.

[1344] Sapiens.

[1345] Sapiens as the first one, Homo Deuce is the second, yeah.

[1346] We mentioned the Black Swan by Talib.

[1347] I think that's a good sort of mind -opener.

[1348] I actually, so there's maybe a more controversial recommendation I could give.

[1349] Great.

[1350] Well, I don't know.

[1351] In some sense, it's so classical it might surprise you.

[1352] But I actually recently read Mortimer Adler's How to Read a Book.

[1353] not so long.

[1354] It was a while ago, but some people hate that book.

[1355] I loved it.

[1356] I think we're in this time right now where, boy, we're just inundated with research papers that you could read on archive with limited peer review and just this wealth of information.

[1357] I don't know, I think the passion of what you can get out of a book, a really good book or a really good paper if you find it, the attitude, the realization that you're only going to find a few that really are worth all your time.

[1358] But then once you find them, you should just dig in and understand it very deeply and it's worth, you know, marking it up and and, you know, having the hard.

[1359] copy writing in the side notes side margins um i think that was really it i read it at the right time where i was just feeling just overwhelmed with really low quality stuff i guess um and similarly uh i'm giving more than three now i'm sorry if i've exceeded my my quota but on that topic just real quick is so basically finding a few companions to keep for the rest of your life in terms of papers and books and so on and those are the ones like not doing um what is it phomo fear missing out constantly trying to update yourself but really deeply making a life journey of studying a particular paper essentially a set of papers yeah i think when you really find something which A book that resonates with you might not be the same book that resonates with me, but when you really find one that resonates with you, I think the dialogue that happens, and that's what I love that Adler was saying, you know, I think Socrates and Plato say, the written word is never going to capture the beauty of dialogue, right?

[1360] But Adler says, no, no. a really good book is a dialogue between you and the author and it crosses time and space and I don't know I think it's a very romantic there's a bunch of like specific advice which you can just gloss over but the romantic view of how to read and really appreciate it is is so good and similarly teaching I I thought a lot about teaching And so Isaac Asimov, great science fiction writer.

[1361] It's also actually spent a lot of his career writing nonfiction, right?

[1362] His memoir is fantastic.

[1363] He was passionate about explaining things, right?

[1364] He wrote all kinds of books on all kinds of topics in science.

[1365] He was known as the great explainer.

[1366] And I do really resonate with his style and just his way of talking about, you know, communicating and explaining to something is really the way that you learn something.

[1367] I think I think about problems very differently because of the way I've been given the opportunity to teach them at MIT.

[1368] We have questions asked, you know, the fear of the lecture, the experience of the lecture and the questions I get and the interactions just forces me to be rock solid on these ideas in a way that I didn't have that.

[1369] I don't know I would be in a different intellectual space.

[1370] Also video, does that scare you that your lectures are online?

[1371] And people like me and sweatpants can sit, sipping coffee and watch you give lectures.

[1372] I think it's great.

[1373] I do think that something's changed right now, which is, you know, right now we're giving lectures over Zoom.

[1374] I mean, giving seminars over Zoom and everything.

[1375] I'm trying to figure out.

[1376] I think it's a new medium.

[1377] Do you think it's...

[1378] I'm trying to figure out how to exploit it.

[1379] Yeah, I've been, I've been quite cynical about human -to -human connection over that medium.

[1380] But I think that's because it hasn't been explored fully.

[1381] And teaching is a different thing.

[1382] Every lecture is a, I'm sorry, every seminar even, I think every talk I give, you know, is an opportunity to give that differently.

[1383] I can I can deliver content directly into your browser.

[1384] You have a WebGL engine right there.

[1385] I can throw 3D content into your browser while you're listening to me, right?

[1386] Yeah.

[1387] And I can assume that you have a, you know, at least a powerful enough laptop or something to watch Zoom while I'm doing that while I'm giving a lecture.

[1388] That's a new communication tool that I didn't have last year, right?

[1389] And I think robotics can potentially benefit a lot from teaching that way.

[1390] we'll see it's going to be an experiment this fall it's interesting thinking a lot about it yeah and also like the the length of lectures or the length of like there's something so like i guarantee you you know it's like 80 % of people who started listening to our conversation are still listening to now which is crazy to me but so there's a there's a patience and interest in long form content but at the same time there's a magic to forcing yourself to condense an idea to as short as possible.

[1391] It's shortest possible like clip.

[1392] It can be a part of a longer thing, but like just like really beautifully condensed an idea.

[1393] There's a lot of opportunity there that's easier to do in remote with, I don't know, with editing too.

[1394] Editing is an interesting thing.

[1395] like what uh you know most professors don't get when they give a lecture you don't get to go back and edit out parts like crisp like crisp it up a little bit uh that's also it can do magic like if you remove like five to ten minutes from an hour lecture it can it can actually it can make something special of a lecture i've uh i've seen that in myself and and in others too because I edit other people's lectures to extract clips.

[1396] It's like there's certain tangents that lose, they're not interesting.

[1397] They're mumbling.

[1398] They're just not, they're not clarifying.

[1399] They're not helpful at all.

[1400] And once you remove them, it's just, I don't know.

[1401] Editing can be magic.

[1402] It takes a lot of time.

[1403] Yeah, it depends, like, what is teaching?

[1404] You have to ask.

[1405] Yeah.

[1406] Because I find the editing process is also beneficial, as for teaching, but also for your own learning.

[1407] I don't know if, have you watched yourself?

[1408] Yeah, sure.

[1409] Have you watched those videos?

[1410] I mean, not all of them.

[1411] It could be painful and to see, like, how to improve.

[1412] So do you find that, I know you segment your podcast, do you think that helps people with the attention span aspect of it?

[1413] Or is it?

[1414] Segment, like sections, like, yeah, we're talking about this topic, whatever.

[1415] No, no, that's just how.

[1416] me it's actually bad so uh and you've been incredible um so i'm learning like i'm afraid of conversation this is even today i'm terrified of talking to you i mean it's it's something i'm trying to remove for myself i there's a guy i mean i've learned from a lot of people but really um there's been a few people who's been inspirational to me in terms of conversation whatever people think of him uh jo rogan has been inspirational to me because as comedians have been to, being able to just have fun and enjoy themselves and lose themselves in conversation.

[1417] That requires you to be a great storyteller, to be able to pull a lot of different pieces of information together, but mostly just to enjoy yourself in conversations, and I'm trying to learn that.

[1418] These notes are, you see me looking down, that's like a safety blanket that I'm trying to let go of more and more.

[1419] Cool.

[1420] so that's that people love just regular conversation that's what they the structure is like whatever uh i would say i would say maybe like 10 to like so there's a bunch of you know there's probably a couple thousand phd students listening to this right now right and they might know what we're talking about but there is somebody i guarantee you right now in russia some kids who's just like who just smoke some weed is sitting back and just enjoying the hell out of this conversation not really understanding he kind of watched some Boston Dynamics videos he's just enjoying it and I salute you sir no but just like there's a so much variety of people that just have curiosity about engineering about sciences about mathematics and and also like I should I mean, enjoying it is one thing, but also often notice it inspires people to, there's a lot of people who are like in their undergraduate studies trying to figure out what, trying to figure out what to pursue.

[1421] And these conversations can really spark the direction of their life.

[1422] And in terms of robotics, I hope it does, because I'm excited about the possibilities of robotics brings.

[1423] On that topic, do you have advice?

[1424] What advice would you give to a young person about life?

[1425] A young person about life or a young person about life in robotics?

[1426] It could be in robotics, it could be in life in general.

[1427] It could be career, it could be relationship advice, it could be running advice.

[1428] That's one of the things I see, like we talk to like 20 -year -olds.

[1429] they're like, how do I, how do I do this thing?

[1430] What do I do?

[1431] If they come up to you, what would you tell them?

[1432] I think it's an interesting time to be a kid these days.

[1433] Everything points to this being sort of a winner -take -all economy and the like.

[1434] I think the people that will really excel, in my opinion, are going to be the ones that can think deeply about problems.

[1435] You have to be able to ask questions agilely and use the internet for everything it's good for and stuff like this.

[1436] And I think a lot of people will develop those skills.

[1437] I think the leaders, thought leaders, robotics leaders, whatever, are going to be the ones that can do more and they can think very deeply and critically.

[1438] And that's a harder thing to learn.

[1439] I think one path to learning that is through mathematics.

[1440] through engineering, I would encourage people to start math early.

[1441] I mean, I didn't really start.

[1442] I mean, I was always in the better math classes that I could take, but I wasn't pursuing super advanced mathematics or anything like that until I got to MIT.

[1443] I think MIT lit me up and really started the life that I'm living now.

[1444] but yeah I really want kids to to dig deep really understand things building things too I mean pull things apart put them back together like that's just such a good way to really understand things and expect it to be a long journey right it's uh you don't have to know everything you're never going to know everything so think deeply and stick with it enjoy the ride but just make sure you're not um yeah just just make sure you're you're you're stopping to think about why things work it's true it's uh it's easy to lose yourself in the in the in the distractions of the world we're overwhelmed with content right now but you have to stop and pick some of it and and really understand yeah on the book point i've read um animal farm by george orwell a ridiculous number of times so for me like that book i don't know if it's a good book in general but for me it connects deeply somehow uh it somehow connects so i was born in the soviet union so it connects to me to the entirety history of the soviet union and to world war two and to the love and hatred and suffering that went on there and the uh the corrupting nature of power and greed and just somehow, I just, that, that book has taught me more about life than like anything else.

[1445] Even though it's just like a silly, like, childlike book about animals, like, I don't know why, it just connects and inspires.

[1446] And the same, there's a few, yeah, there's a few technical books, too, and algorithms that just, yeah, you return too often.

[1447] Right.

[1448] I'm, I'm with you.

[1449] Yeah, there's, I don't, and I've been losing that because of the, internet.

[1450] I've been like going on archive and blog posts and GitHub and the new thing of you lose your ability to really master an idea.

[1451] Right.

[1452] Wow.

[1453] Exactly right.

[1454] What's a fond memory from childhood when baby Russ Tedrick?

[1455] Well I guess I just said that at least my current life begins, began when I got to MIT, if I have to go farther than that.

[1456] Yeah, what was, was there a life before MIT?

[1457] Oh, absolutely.

[1458] But, but let me actually tell you what happened when I first got to MIT, because that, I think might be relevant here.

[1459] But I, you know, I had taken a computer engineering degree at Michigan.

[1460] I enjoyed it immensely, learned a bunch of stuff.

[1461] I was, I liked computers.

[1462] I've liked programming.

[1463] But when I did get to MIT and started working with Sebastian Sung, theoretical physicist, computational neuroscientist, the culture here was just different.

[1464] It demanded more of me, certainly mathematically and in the critical thinking.

[1465] And I remember the day that I borrowed one of the books from my advisor's office and walked down to the Charles River and was like, I'm getting my butt kicked, you know?

[1466] And I think that's going to happen to everybody who's doing this kind of stuff, right?

[1467] I think I expected you to ask me the meaning of life.

[1468] You know, I think that the, somehow I think that's, that's got to be part of it.

[1469] Doing hard things?

[1470] Yeah.

[1471] Did you consider quitting at any point?

[1472] Did you consider this isn't for me?

[1473] No, never that.

[1474] I mean, I was, I was working hard, but I was loving it.

[1475] I mean, there's, I think there's this magical thing where you, you know, I'm lucky to surround myself with people that basically, almost every day, I'll, I'll see something, I'll be told something or something that I realize, wow, I don't understand that.

[1476] And if I could just understand that, there's, there's something else to learn that if I could just learn that thing, I would connect another piece of the puzzle.

[1477] And, you know, I think that is just such an important aspect and being willing to understand what you can and can't do and loving the journey of going and learning those other things.

[1478] I think that's the best part.

[1479] I don't think there's a better way to end it.

[1480] Russ, you've been an inspiration to me since I showed up at MIT.

[1481] Your work has been an inspiration to the world.

[1482] This conversation was amazing.

[1483] I can't wait to see what you do next with robotics, home robots.

[1484] I hope to see your work in my home one day.

[1485] Thanks so much for talking today.

[1486] It's been awesome.

[1487] Cheers.

[1488] Thanks for listening to this conversation with Ross Tedrick.

[1489] And thank you to our sponsors, MagicSpoon, Cereal, BetterHelp, and ExpressVPN.

[1490] Please consider supporting this podcast by going to magic spoon .com slash Lex and using code Lex at checkout.

[1491] I'm going to BetterHelp .com slash Lex.

[1492] and signing up at expressvpn .com slash LexPod.

[1493] Click the links, buy the stuff, get the discount.

[1494] It really is the best way to support this podcast.

[1495] If you enjoy this thing, subscribe on YouTube, review it with five stars on Apple Podcast, support it on Patreon, or connect with me on Twitter at Lex Friedman, spelled somehow without the E, just F -R -I -D -M -A -N.

[1496] And now, let me leave you with some words from Neil deGrasse Tyson, talking about robots in space and the emphasis we humans put on human -based space exploration.

[1497] Robots are important.

[1498] If I don my pure scientist hat, I would say just send robots.

[1499] I'll stay down here and get the data.

[1500] But nobody's ever given a parade for a robot.

[1501] Nobody's ever named a high school after a robot.

[1502] So when I don't my public educator hat, I have to recognize the elements of exploration that excite people.

[1503] It's not only the discoveries and the beautiful photos that come down from the heavens.

[1504] It's the vicarious participation in discovery itself.

[1505] Thank you for listening and hope to see you next time.