The Quest for Natural-Looking Gait: A 15-Year Journey with Boston Dynamics' Atlas Robot

Our goal was to achieve a natural-looking gait, and it proved to be surprisingly challenging. We did manage to build an early machine, which we called the PETMAN prototype. This precursor to the PETMAN robot had a nice-looking gait, where the leg would stick out and perform heel strike before rolling onto the toe, ensuring that the foot never landed flat. However, even with this initial success, it was hard to get the robot to walk naturally when it fully extended its leg.

Even after years of development, we didn't quite achieve the natural walking motion we had hoped for until around last year. The team working on our newer generation of Atlas, a humanoid robot, developed new techniques for creating a walking-control algorithm as part of their work. These techniques led to a byproduct of natural-looking motion, which was not a primary goal but rather an unexpected outcome.

The journey towards achieving natural-looking gait took around 15 years, from the PETMAN prototype in 2008 to the last year. This highlights the complexity and challenges involved in creating robots that can mimic human movement with ease.

Boston Dynamics: A Legendary Robotics Company

Boston Dynamics is a legendary robotics company that has been creating some of the most elegant, dexterous, and simply amazing robots ever built over the past 30 years. The company's humanoid robot Atlas and robot dog Spot are just two examples of their impressive work, having been featured on the Internet for their incredible abilities such as dancing, doing backflips, opening doors, or throwing around heavy objects.

Robert Playter, CEO of Boston Dynamics, has led both the development of the company's humanoid robots and its physics-based simulation software. He has been with the company since its inception and was even part of the team at MIT where he received his PhD in aeronautical engineering in 1994.

The MIT Leg Lab, where Robert received his PhD, played an essential role in shaping Boston Dynamics into what it is today. His work on robot gymnastics as part of his thesis included programming a bipedal robot to perform the world's first 3D robotic somersault. This showcases his expertise and passion for robotics, which has been instrumental in driving the company's success.

A Conversation with Robert Playter

This conversation with Robert Playter was a big honor and pleasure, and he expressed hope to collaborate on great work with these robots in the years to come. The Lex Fridman podcast is proud to have had this opportunity to speak with him.

Love and Robots: The Fascination with Movement

Robert Playter's love for robotics began when he visited MIT looking for a place to get his PhD. He was drawn to the field because of its deep fascination with movement. This attraction has been a driving force behind his work, which has led to some remarkable achievements in the world of robotics.

A Conversation Starter: When Did You First Fall in Love with Robotics?

When asked when he first fell in love with robotics, Robert Playter shared that it happened during his visit to MIT. He was visiting the department looking for a place to conduct laboratory work and found himself drawn to the field due to its fascination with movement.

"WEBVTTKind: captionsLanguage: en- And so our goal wasa natural-looking gait.It was surprisingly hardto get that to work.But we did build an early machine.We called it PETMAN prototype.It was the prototypebefore the PETMAN robot,and it had a really nice-looking gaitwhere, you know, itwould stick the leg out.It would do heel strike firstbefore it rolled onto the toe,so you didn't land with a flat foot.You extended your leg alittle bit, but even then,it was hard to get the robot to walkwhere, when you were walking,that it fully extended its legand getting that all to workwell took such a long time.In fact, I probably didn't really seethe nice, natural walking that I expectedout of our humanoidsuntil maybe last year.And the team was developingon our newer generationof Atlas, you know, some new techniquesfor developing awalking-control algorithm.And they got thatnatural-looking motion as sortof a byproduct of just a different processthey were applying todeveloping the control.So, that probably took 15years, 10 to 15 years to sortof get that from, you know,the PETMAN prototype was probably in 2008,and what was it, 2022, (laughs) last yearthat I think I saw good walking on Atlas.(dramatic music)- The following is aconversation with Robert Playter,CEO of Boston Dynamics, alegendary robotics companythat, over 30 years, has createdsome of the most elegant,dextrous, and simplyamazing robots ever built,including the humanoid robotAtlas and the robot dog Spot,one or both of whom you'veprobably seen on the Internet,either dancing, doingbackflips, opening doors,or throwing around heavy objects.Robert has led both the developmentof Boston Dynamics humanoid robotsand their physics-basedsimulation software.He has been with the companyfrom the very beginning,including its roots at MIT,where he received his PhDin aeronautical engineering.This was in 1994 at thelegendary MIT Leg Lab.He wrote his PhD thesison robot gymnasticsas part of which he programmeda bipedal robot to dothe world's first 3D robotic somersault.Robert is a great engineer,roboticist, and leader,and Boston Dynamics, tome as a roboticist, isa truly inspiring company.This conversation was abig honor and pleasure,and I hope to do a lot of great workwith these robots in the years to come.This is the Lex Fridman podcast.To support it, pleasecheck out our sponsorsin the description.And now, dear friends,here's Robert Playter.When did you first fallin love with robotics?(Lex laughs)Let's start with love and robots.- Well, love is relevant because I thinkthe fascination, the deep fascination isreally about movement, andI was visiting MIT lookingfor a place to get a PhD,and I wanted to do some laboratory work.And one of my professors inthe aero department said,\"Go see this guy Marc Raibertdown in the basement of the AI lab.\"And so I walked down there and saw him.He showed me his robots,and he showed me thisrobot doing a somersault.(Lex laughs)And I just immediatelywent, \"Whoa,\" you know.- Yeah.- \"Robots can do that?\"And because of my owninterest in gymnastics,there was, like, thisimmediate connection,and, you know, I wasin an aeroastro degreebecause, you know, flight and movement wasall so fascinating to me.And then it turned outthat, you know, roboticshad this big challenge.How do you balance?How do you build a legged robotthat can really get around?That was a fascination,and it still exists today.We're still working onperfecting motion in robots.- What about the elegance and the beautyof the movement itself?Is there something maybegrounded in your appreciationof movement from your gymnastics days?Was there something you justfundamentally appreciatedabout the elegance and beauty of movement?- You know, we had this conceptin gymnastics of letting yourbody do what it wanted to do.When you get really good at gymnastics,part of what you're doingis putting your bodyinto a position where the physicsand the body's inertia andmomentum will kinda push youin the right direction in avery natural and organic way.And the thing that Marcwas doing, you know,in the basement of that laboratorywas trying to figure outhow to build machines to takeadvantage of those ideas.How do you build somethingso that the physicsof the machine justkind of inherently wantsto do what it wants to do?And he was building these springypogo-stick type, you know.His first cut at leggedlocomotion was a pogo stickwhere it's bouncing, andthere's a spring mass systemthat's oscillating, has its own sortof natural frequencythere and sort of figuringout how to augment those naturalphysics with also intent.How do you then controlthat but not overpower it?It's that coordination that Ithink creates real potential.We could call it beauty, you know.You could call it, I don't know, synergy.People have different words for it.But I think that that wasinherent from the beginning.That was clear to me that that's partof what Marc was trying to do.He asked me to do thatin my research work.So, you know, that's where it got going.- So, part of the thing thatI think I'm calling eleganceand beauty in this case, which was there,even with the pogo stickis maybe the efficiency,so letting the body dowhat it wants to do,trying to discover the efficient movement.- It's definitely more efficient.It also becomes easierto control in its own waybecause the physics are solvingsome of the problem itself.It's not like you have todo all this calculationand overpower the physics.The physics naturally, inherently wantto do the right thing.There can even be, youknow, feedback mechanisms,stabilizing mechanismsthat occur simply by virtueof the physics of the body.And it's, you know,not all in the computeror not even all in yourmind as a person (laughs).And there's somethinginteresting in that melding.- You were with Marc formany, many, many years,but you were there inthis kinda legendary spaceof Leg Lab and MIT inthe basement (laughs).All great things happen in the basement.(Robert laughs)Is there some memories fromthat time that you have?Because it's such cutting-edge workin robotics and artificial intelligence.- The memories, the distinctivelessons, I would sayI learned in that time periodand that I think Marc wasa great teacher of wasit's okay to pursue yourinterests, your curiosity,do something because you love it.You'll do it a lot better if you love it.That is a lasting lessonthat I think we applyat the company still andreally is a core value.- So, the interesting thing is,with people like Russ Tedrake and others,like, the students that workat those robotics labs are, like, someof the happiest people I've ever met.I don't know what that is. (laughs)I meet a lot of PhD students.A lot of them are kind of broken(laughing) by the wearand tear of the process,but roboticists are, whilethey work extremely hardand work long hours,there's a happiness there.The only other group of peopleI've met like that are peoplethat skydive a lot.(both laughing)For some reason, there's adeep, fulfilling happiness maybefrom, like, a long period of struggleto get a thing to work, and it works,and there's a magic to it.I don't know exactly 'causeit's so fundamentally hands-on,and you're bringing a thing to life.I don't know what itis, but they're happy.- You know, our attrition atthe company is really low.People come, and they love the pursuit.And I think part of that isthat there's perhaps anatural connection to it.It's a little bit easier toconnect when you have a robotthat's moving around in theworld, and part of your goal isto make it move around in the world.You can identify with that.This is one of the unique thingsabout the kinds ofrobots we're building isthis physical interaction letsyou perhaps identify with it.So, I think that is a source of happiness.I don't think it's unique to robotics.I think anybody also who is just pursuingsomething they love, it'seasier to work hard at itand be good at it, and noteverybody gets to find that.I do feel lucky in that way.And I think we're lucky as an organizationthat we've been able tobuild a business around thisand that keeps people engaged.- So, if it's all right,let's linger on Marcfor a little bit longer, Marc Raibert.So, he's a legend.He's a legendary engineer and roboticist.What have you learned aboutlife, about robotics from Marcthrough all the many yearsyou've worked with him?- I think the most importantlesson, which was, you know,have the courage of your convictionsand do what you think is interesting.Be willing to try to findbig, big problems to go after.And at the time, youknow, legged locomotion,especially in a dynamicmachine, nobody had solved it.And that felt like amulti-decade problem to go after.And so, you know, have thecourage to go after thatbecause you're interested.Don't worry if it's gonna make money.You know, that's been a theme.That's really probably themost important lesson I thinkthat I got from Marc.- How crazy is the effortof doing legged roboticsat that time, especially?- You know, Marc got somestuff to work startingfrom simple ideas.So, maybe the other,another important ideathat has really become avalue of the company istry to simplify a thingto the core essence.While, you know, Marc wasshowing videos of animals runningacross the Savannah or climbing mountains,what he started with was a pogo stickbecause he was trying toreduce the problem to somethingthat was manageable, andgetting the pogo stickto balance had in itthe fundamental problemsthat, if we solved those, youcould eventually extrapolateto something that gallopedlike a horse, and so lookfor those simplifying principles.- How tough is the jobof simplifying a robot?- So, I'd say, in the earlydays, the thing that madethe researchers at BostonDynamics special isthat we worked on figuring outwhat that central principle wasand then building software or machinesaround that principle,and that was not easyin the early days.And it took real expertisein understanding the dynamicsof motion and feedback-controlprinciples, how to build,you know, with the computers at the time,how to build a feedback-control algorithmthat was simple enough thatit could run in real timeat 1,000 hertz and actuallyget that machine to work.And that was not somethingeverybody was doing,you know, at that time.Now, the world's changing now,and I think the approachesto controlling robots are going to change,and they're going to becomemore broadly available.But at the time, there weren't many groupswho could really sort ofwork at that principled levelwith both the software andmake the hardware work.And I'll say one other thingabout you were sort of talkingabout what are the special things.The other thing was it's goodto break stuff, you know.You know, use the robots,break them, repair them,you know, fix and repeat,(laughs) test, fix, and repeat.And that's also a coreprinciple that has become partof the company, and it letsyou be fearless in your work.Too often, if you are workingwith a very expensive robot,maybe one that youbought from somebody elseor that you don't know how to fix,then you treat it with kid gloves,and you can't actually make progress.You have to be able to break something.And so, I think that'sbeen a principle as well.- So, just to linger onthat, psychologically,how do you deal with that?'Cause I remember I built a RC car.It had some custom stufflike a computer on itand all that kind of stuff, camerasand because I didn't sleep much,the code I wrote had an issuewhere it didn't stop the car,and the car got confused and at full speedat, like, 20, 25 miles anhour, it slammed into a wall.And I just remember sittingthere alone in a deep sadness,sort of full of regret,I think, almost anger,but also, like, sadnessbecause you think about,well, these robots, especiallyfor autonomous vehicles,like, you should be takingsafety very seriouslyeven in these kinds of things,but just no good feelings.It made me more afraidprobably to do these kindof experiments in the future.Perhaps the right way tohave seen that is positively.Like, it's too-- It depends if youcould have built that caror just gotten another one, right?That would've been the approach.I remember when I got to grad school,you know, I got some trainingabout operating a latheand a mill up in the machine shop,and I could start to make my own parts.And I remember breaking somepiece of equipment in the laband then realizing 'causemaybe this was a unique part,and I couldn't go buy it, and I realized,\"Oh, I can just go make it.\"That was an enabling feeling.- Yeah.- Then, you're not afraid.It might take time.It might take more work than you thoughtit was gonna be requiredto get this thing done,but you can just go make it.And that's freeing in away that nothing else is.- You mentioned the feedbackcontrol, the dynamics,sorry for the romantic question,but in the early days andeven now, is the dynamics,probably more appropriatefor the early days,is it more art or science?- There's a lot of sciencearound it, and trying to develop,you know, scientific principlesthat let you extrapolatefrom, like, one legged machine to another,you know, develop a core set of principleslike a spring-mass bouncingsystem and then figure out howto apply that from a one-legged machineto a two- or a four-legged machine.Those principles are really importantand were definitely acore part of our work.There's also, you know, when we startedto pursue humanoid robots,there was so much complexityin that machine that, youknow, one of the benefitsof the humanoid form isyou have some intuitionabout how it shouldlook while it's moving.And that's a littlebit of an art, I think,or maybe it's justtapping into a knowledgethat you have deep inyour body and then tryingto express that in the machine,but that's an intuitionthat's a little bit more on the art side.Maybe it predates your knowledge.Before you have the knowledgeof how to control it,you try to work throughthe art channel. (laughs)- Yeah.- And humanoids sortof make that available to you.If it had been a different shape,maybe you wouldn't have hadthe same intuition about it.- Yeah, so your knowledge about movingthrough the world is notmade explicit to you.That's why it's art.- Yeah, it might be hard toactually articulate exactly.(laughing) You know?- Yeah.- And being a competitive athlete,there's something about seeing a movement.You know, a coach, oneof the greatest strengthsa coach has is beingable to see, you know,some little change inwhat the athlete is doingand then being able toarticulate that to the athlete,you know, and then maybeeven trying to say,\"And you should try to feel this.\"So, there's something just in seeing,and again, you know, sometimesit's hard to articulatewhat it is you're seeing, butjust perceiving the motionat a rate that is, again,sometimes hard to put into words.- Yeah, I wonder how it ispossible to achieve sortof truly elegant movement.You have a movie like \"Ex Machina.\"I'm not sure if you've seen it,but the main actress inthat who plays the AI robotI think is a ballerina.I mean, just the natural eleganceand the, I don't know,eloquence of movement, (laughs)it looks efficient and easy,and just it looks right.It looks beautiful.- It looks right issort of the key, yeah?- And then, you look at,especially early robots,I mean, they're so cautiousin the way they movethat it's not thecaution that looks wrong.It's something about themovement that looks wrongthat feels like it's veryinefficient, unnecessarily so.And it's hard to putthat into words exactly.- We think that, and part of the reasonwhy people are attractedto the machines we build isbecause the inherent dynamicsof movement are closer to rightbecause we try to use,you know, walking gaits,or we build a machine around this gaitwhere you're trying to workwith the dynamics of the machineinstead of to stop them.You know, some of the earlywalking machines, you know,you're essentially,you're really trying hardto not let them fall over,and so you're always stoppingthe tipping motion, you know.And sort of the insightof dynamic stabilityin a legged machine is to gowith it, you know, (laughs)let the tipping happen.You know, let yourself fall,but then catch yourselfwith that next foot.And there's somethingabout getting those physicsto be expressed in themachine that people interpretas lifelike, or elegant,or just natural looking.And so, I think if youget the physics right,it also ends up beingmore efficient, likely.There's a benefit that it probably endsup being more stable in the long run.You know, it could walkstably over a wider rangeof conditions, and it's more beautifuland attractive at the same time.- So, how hard is it to getthe humanoid robot Atlasto do some of the things thatit's recently been doing?Let's forget the flips and all of that.Let's just look at the running.Maybe you can correct me,but there's something about running.I mean, that's not careful at all.That's you're falling forward.You're jumping forward and are falling.So, (laughing) how hardis it to get that right?- Our first humanoid, we needed to delivernatural-looking walking, you know.We took a contract from the army.They wanted a robot thatcould walk naturally.They wanted to put a suit on the robotand be able to test itin a gas environment.And so, they wanted themotion to be natural.And so, our goal was anatural-looking gait.It was surprisingly hardto get that to work.But we did build an early machine.We called it PETMAN prototype.It was the prototypebefore the PETMAN robot,and it had a really nice-looking gaitwhere, you know, itwould stick the leg out.It would do heel strike firstbefore it rolled onto the toe,so you didn't land with a flat foot.You extended your leg a little bit,but even then it was hardto get the robot to walkwhere, when you were walking,that it fully extended its legand essentially landed on an extended leg.And if you watch closely how you walk,you probably land on an extended leg,but then you immediately flex your kneeas you start to make that contact,and getting that all to workwell took such a long time.In fact, I probably didn't really seethe nice, natural walking that I expectedout of our humanoidsuntil maybe last year.And the team was developingon our newer generationof Atlas, you know, some new techniquesfor developing awalking-control algorithm.And they got thatnatural-looking motion as sortof a byproduct of just a different processthey were applying todeveloping the control.So, that probably took 15years, 10 to 15 years to sortof get that from, you know,the PETMAN prototype was probablyin 2008, and what was it,2022, (laughs) last yearthat I think I saw good walking on Atlas.- If you could just, like, linger on it,what are some challengesof getting good walking?So, is this partially,like, a hardware, like, actuator problem?Is it the control?Is it the artisticelement of just observingthe whole system operating indifferent conditions together?I mean, is there somekind of interesting quirksor challenges you can speakto, like the heel strikeor all this kind of stuff?- Yeah, so oneof the things that makes,like, this straight lega challenge is you're sortof up against a singularity,a mathematical singularitywhere, you know,when when your leg is fully extended,it can't go further theother direction, right?You can only move in onedirection, and that makes allof the calculations aroundhow to produce torquesat that joint or positionsmakes it more complicated.And so, (laughs) havingall of the mathematicsso it can deal with thesesingular configurations is oneof many (laughs) challenges that we face.And I'd say, you know, inthose earlier days, again,we were working with thesereally simplified models.So, we're trying to boil all the physicsof the complex human bodyinto a simpler subsystemthat we can more easilydescribe in mathematics.And sometimes those simplersubsystems don't have allof that complexity of thestraight leg built into them.And so, what's happenedmore recently is we're ableto apply techniques thatlet us take the full physicsof the robot into accountand deal with some ofthose strange situationslike the straight leg.- So, is there a fundamentalchallenge here that it's,maybe you can correct me,but is it underactuated?Are you falling?- Underactuated is the right word, right?You can't push the robot inany direction you want to.- Yeah.- Right?And so, that is one of the hard problemsof legged locomotion.- And you have to do thatfor natural movement?- It's not necessarilyrequired for natural movement.It's just required, you know,we don't have, you know,a gravity force that you canhook yourself onto to applyan external force inthe direction you wantat all times, right?The only external forcesare being mediatedthrough your feet, and howthey get mediated dependon how you place your feet,and you know, you can't just,you know, God's handcan't reach down and pushin any direction you want,(laughs) you know, so.- Is there some extrachallenge to the factthat Atlas is such a big robot?- There is.The humanoid form isattractive in many ways,but it's also a challenge in many ways.You have this big upperbody that has a lotof mass and inertia, andthrowing that inertiaaround increases the complexityof maintaining balance.And as soon as you pick upsomething heavy in your arms,you've made that problem even harder.And so, in the early work in the Leg Laband in the early days atthe company, you know,we were pursuing these quadruped robots,which had a kind ofbuilt-in simplification.You had this big rigid bodyand then really light legs.So, when you swing the legs,the leg motion didn't impactthe body motion very much.All the mass and inertia was in the body,but when you have thehumanoid, that doesn't work.You have big heavy legs.You swing the legs.It affects everything else.(Lex laughs)And so, dealing with all ofthat interaction does makethe humanoid a much morecomplicated platform.- And I also saw that at leastrecently you've been doingmore explicit modelingof the stuff you pick up.- Yeah, yeah.- Which is (laughs) really interesting.So, you have to, what, model the shape,the weight distribution.I don't know, like, youhave to, like, include thatas part of the modeling,as part of the planning'cause okay, so for peoplewho don't know, so Atlas,at least in, like, a recentvideo, like, throws a heavy bag,throws a (laughing) bunch of-- Yeah.- stuff.So, what's involved in pickingup a thing, a heavy thing?And when that thing is a bunchof different non-standard things,I think it also picked up like a barbelland to be able to throw in some cases,what are some interestingchallenges there?- So, we were definitelytrying to show that the robotand the techniques we'reapplying to Atlas let us dealwith heavy things in the world.Because if the robot's gonna be useful,it's actually gotta move stuff around.And that needs to be significant stuffthat's an appreciable portionof the body weight of the robot.And we also think this differentiates usfrom the other humanoid robot activitiesthat you're seeing out there.Mostly, they're not picking stuff up yet,not heavy stuff anyway.But just like you or me, you know,you need to anticipate that moment.You know, you're reachingout to pick something up,and as soon as you pick it up,your center of mass is gonna shift.And if you're gonna, youknow, turn in a circle,you have to take thatinertia into account.And if you're gonnathrow a thing, you know,all of that has to be sort of includedin the model of what you're trying to do.So, the robot needs to havesome idea or expectationof what that weight is andsort of predict, you know,think a couple of seconds ahead,\"How do I manage now my bodyplus this big heavy thingtogether (laughs) and stillmaintain balance, right?\"And so, that's a big change for us,and I think the tools we'vebuilt are really allowingthat to happen quickly now.Some of those motions that yousaw in that most recent videowe were able to createin a matter of days.It used to be that it tooksix months to do anything new,you know, on the robot, andnow we're starting to developthe tools that let us dothat in a matter of days.And so, we think that's really exciting.That means that the abilityto create new behaviorsfor the robot is gonnabe a quicker process.- So, being able toexplicitly model new thingsthat it might need to pickup, new types of things?- And you know, to some degree,you don't wanna have topay too much attentionto each specific thing, right?There's sort of a generalization here.Obviously, when you grab a thing,you have to conform yourhand, your end effectorto the surface of that shape,but once it's in your hands,it's probably just the massand inertia that matter,and the shape may not be as important.- Yeah.- And so, you know, in someways you wanna pay attentionto that detailed shape, and in others,you wanna generalize it and say,\"Well, all I really care about isthe center of mass of this thing,especially if I'm gonna throwit up on that scaffolding.\"- And it's easier if the body is rigid.What if there's some, doesn't it throw,like, a sandbag type thing?- That tool bag, you know-- Tool bag.- had loose stuff init, so it managed that.There are harder thingsthat we haven't done yet.You know, we could havehad a big jointed thingor, I don't know, a bunchof loose wire or rope.- What about carrying another robot?How 'bout that? (laughing)- Yeah, we haven't done that yet.- Carry Spot.- I guess we did a little bit of a,we did a little skit around Christmaswhere we had two Spotsholding up another Spotthat was trying to put,you know, a bow on a tree.So, I guess we're doing thatin a small way. (laughing)- Okay, that's pretty good.Let me ask the all-important question.Do you know how much Atlas can curl?(Robert laughing drown out Lex speaking)(Lex laughs)I mean, you know, for ushumans, that's really oneof the most fundamental questionsyou can ask another human being,curl, bench, et cetera.(Robert laughs)- It probably can't curlas much as we can yet,but a metric that Ithink is interesting is,you know, another way oflooking at that strength is,you know, the box jump.So, how high of a box can you jump onto?- Question.- And Atlas, I don'tknow the exact height.It was probably a meterhigh or something like that.It was a pretty pretty talljump that Atlas was ableto manage when we last tried to do this.And I have video of mychief technical officerdoing the same jump, and hereally struggled, you know,to get-- Oh, the human?- The human getting allthe way on top of this box.But then, you know,Atlas was able to do it.We're now thinking about thenext generation of Atlas,and we're probably gonna be in the realmof a person can't do it, youknow, with the next generation.The robots, the actuatorsare gonna get strongerwhere it really is thecase that at least someof these joints, some of thesemotions will be stronger.- And to understand how high it can jump,you probably had to doquite a bit of testing.- Oh, yeah, and there'slots of videos of it tryingand failing, and you know,that's all, you know,we don't always release those videos,but they're a lot offun to look at. (laughs)- So, we'll talk a little bit about that.But can you talk to the jumping?'Cause you talked about the walking,and it took a long time, many, many yearsto get the walking to be natural,but there's also really natural-looking,robust, resilient jumping.How hard is it to do the jumping?- Well, again, this stuffhas really evolved rapidlyin the last few years.You know, the first time wedid a somersault, you know,there was a lot of kindof manual iteration.What is the trajectory?You know, how hard do you throw?In fact, in these early days,when I'd see early experimentsthat the team was doing,I might make suggestions abouthow to change the technique,again, kind of borrowingfrom my own intuitionabout how backflips work.But frankly they don't need that anymore.So, in the early days,you had to iterate kindof in almost a manual way tryingto change these trajectoriesof the arms or the legsto try to get, you know, asuccessful backflip to happen.But more recently, we're runningthese model-predictive control techniqueswhere the robot essentiallycan think in advancefor the next second or twoabout how its motion is goingto transpire, and you can, you know, solvefor optimal trajectoriesto get from A to B.So, this is happening ina much more natural way,and we're really seeingan acceleration happenin the development of these behaviors,again, partly due to theseoptimization techniques,sometimes learning techniques,so it's hard in that there's alot of mathematics behind it,but we're figuring that out.- So, you can domodel-predictive control for,I mean, I don't evenunderstand what that looks likewhen the entire robot is in the airflying and doing a back (laughs).- Yeah, well-- I mean. (laughs)- But that's the cool part, right?So, you know, the physics,we can calculate physicspretty well using, you know,Newton's laws about how it'sgoing to evolve over timeand you know, the sick trick,which was a front somersaultwith a half twist isa good example, right?You saw the robot on variousversions of that trick.I've seen it land indifferent configurations,and it still manages to stabilizeitself, and so, you know,what this model-predictivecontrol means is, again,in real time, the robot isprojecting ahead, you know,a second into the future andsort of exploring options.And if I move my arm alittle bit more this way,how is that gonna affect the outcome?And so, it can do thesecalculations, many of them,you know, and basically solvefor where, you know, given where I am now,maybe I took off a little bit screwyfrom how I had planned, I can adjust.- So, you're adjusting in the airfor the landing.- Adjust on the fly.So, the model-predictivecontrol lets you adjuston the fly, and of course, I think this iswhat, you know, people adapt as well.When we do it, even a gymnastics trick,we try to set it up so it'sclose to the same every time.But we figured out how to dosome adjustment on the fly,and now we're starting to figure outthat the robots can dothis adjustment on the flyas well using these techniques.- In the air.I mean, it just feels,from a robotics perspective, just surreal.- You talked about underactuated, right?- Yes.- So, when you're-- That's totally true.- When you're in the air,there's some things youcan't change, right?You can't change the momentumwhile it's in the air'cause you can't apply anexternal force, a torque,and so the momentum isn't gonna change.So, how do you work within the constraintof that fixed momentum tostill get from A to B (laughs)where you wanna be?- That's really (laughing) underactuated.(Robert laughs)You're in the air.I mean, you become a dronefor a brief moment in time.No, you're not even a (laughing)drone 'cause you can't-- Can't hover.- You can't hover.You can't.- You're gonna impact soon.Be ready. (laughs)- Yeah.Have you considered likea hover type thing or no?No?- No.- It's too much weight?- No.(Lex laughing)- I mean, it's justincredible and just evento have the guts to try abackflip with such a large body.That's wild.(Robert laughs)But, like how-- We definitely broke afew robots trying that.- (laughing) Yeah.(Robert laughs)- But that's where thebuild it, break it, fix it,you know, strategy comes in.You gotta be willing to break.And what ends up happening isby breaking the robot repeatedly,you find the weak points,and then you end up redesigning itso it doesn't break so easilynext time, you know. (laughs)- Through the breakingprocess you learn a lot,like, a lot of lessons,and you keep improvingnot just how to make the backflip work,but everything just-- Yeah.And how to build the machine better.- Yeah.- Yeah.- I mean, is there somethingabout just the gutsto come up with an ideaof saying, \"You know what?Let's try (laughing) tomake it to a backflip\"?- Well, I think thecourage to do a backflipin the first place andto not worry too muchabout the ridicule of somebody saying,\"Why the heck are you doingbackflips with robots?\"- Sure.- Because a lot of peoplehave asked that, you know.(Lex laughs)(laughing) \"Why are you doing this?\"- Why go to the moon(Robert laughs)in this decade and dothe other things, JFK?(Robert laughs)Not because it's easy, because it's hard.- Yeah, exactly. (laughs)(Lex laughs)- Don't ask questions.Okay, so the jumping, I mean,there's a lot of incredible stuff.If we can just rewind a little bitto the DARPA RoboticsChallenge in 2015, I think,which was, for people who aren't familiarwith the DARPA challenges, it was firstwith autonomous vehicles,and there's a lotof interesting challenges around that.And the DARPA Robotics Challenge waswhen humanoid robots weretasked to do all kindsof, you know, manipulation, walking-- Driving a vehicle.- driving a car,all these kinds of challengeswith, if I remember correctly,sort of some slight capabilityto communicate with humans,but the communication was very poor.So, basically it has to bealmost entirely autonomous.- It could have periodswhere the communicationwas entirely interrupted,and the robot had to be able to proceed.- Yeah.- But you could providesome high-level guidanceto the robot, basicallylow-bandwidth communications-- Yeah- to steer it.- I watched that challengewith kind of tears in my eyeseating popcorn with-- Us, too.(both laughing)- But I wasn't personallylosing, you know,hundreds of thousands, millions of dollarsand many years of incredible, hard workby some of the most brilliantroboticists in the world.So, that was why the tragic,that's why tears came.(Robert laughs)So, anyway, just lookingback to that time,what have you learnedfrom that experience?And maybe if you coulddescribe what it wassort of the setup forpeople who haven't seen it.- Well, so there was a contestwhere a bunch of differentrobots were askedto do a series of tasks, someof those that you mentioned,drive a vehicle, get out, open a door,go identify a valve, shut a valve,use a tool to maybecut a hole in a surfaceand then crawl over some stairsand maybe some rough terrain.So, the idea was havea general-purpose robotthat could do lots of different things,had to be mobility, andmanipulation, on-board perception.And there was a contest,which DARPA likes at thetime, was running sortof follow-on to the grand challenge,which was, \"Let's try topush vehicle autonomy along.\"Right?They encouraged peopleto build autonomous cars.So, they were trying to basicallypush an industry forward.Our role in this was to build a humanoid.At the time, it was our sortof first-generation Atlas robot,and we built maybe 10 of them,I don't remember the exact number.And DARPA distributed thoseto various teams that sortof won a contest, showed thatthey could, you know, programthese robots and then use themto compete against each other,and then other robotswere introduced as well.Some teams built their own robots.Carnegie Mellon, for example,built their own robot.And all these robots competedto see who could sort of getthrough this maze the fastest.And again, I think the purpose was to kindof push the whole industry forward.We provided the robot andsome baseline software,but we didn't actuallycompete as a participantwhere we were trying to,you know, drive the robotthrough this maze.We were just trying tosupport the other teams.It was humbling becauseit was really a hard task.And honestly, the tears werebecause, mostly, the robotsdidn't do it. (laughs)You know, they fell down,you know, repeatedly.It was hard to get through this contest.Some did, and you know,they were rewarded and won.But it was humblingbecause of just how hard,these tasks weren't all that hard.A person could have done it very easily,but it was really hard to getthe robots to do it, you know.And the-- The general nature of it,the variety of it.- The variety.- And also, I don't knowif the tasks were (sighs)sort of the task inthemselves help us understandwhat is difficult and what is not.I don't know if that was obviousbefore the contest was designed,so you kind of tried to figure that out.And I think Atlas is reallya general robot platform,and it's perhaps not bestsuited for the specific tasksof that contest, like just for example,probably the hardest task isnot the driving of the carbut getting in and out of the car.(Robert laughs)And Atlas probably is,you know, if you wereto design a robot that canget into the car easilyand get out easily, youprobably would not make Atlasthat particular car.- Yeah, the robot was a little bit big-- Yeah.- to get inand out of that car, right?- It doesn't fit, yeah.- This is the curse ofa general-purpose robot,that they're not perfect at any one thing,but they might be able todo a wide variety of things.And that is the goalat the end of the day.You know, I think we all wannabuild general-purpose robotsthat can be used for lotsof different activities,but it's hard, and the wisdomin building successful robotsup until this point have been,\"Go build a robot for a specific task,and it'll do it very well.\"And as long as youcontrol that environment,it'll operate perfectly,but robots need to be ableto deal with uncertainty.If they're gonna be usefulto us in the future,they need to be able to dealwith unexpected situations.And that's sort of the goalof a general-purposeor multipurpose robot.And that's just darn hard.And so, yeah, there was thesecurious little failures.Like, I remember a robot, you know,the first time you startto try to push on the worldwith a robot, you forgetthat the world pushes backand will push you over (laughs)if you're not ready for it.And the robot, you know,reached to grab the door handle.I think it missed thegrasp of the door handle,was expecting that its handwas on the door handle,and so when it tried to turn the knob,it just threw itself over.It didn't realize, \"Oh, Ihad missed the door handle.I was expecting a forceback from the door.It wasn't there, andthen I lost my balance.\"And so, these little simple thingsthat you and I wouldtake totally for grantedand deal with, (laughs)the robots don't knowhow to deal with yet, andso you have to start to dealwith all of those circumstances. (laughs)- Well, I think a lotof us experience thiseven when sober but drunk, too.Sort of, you pick up athing and expect it to be,what is it, heavy, andit turns out to be light.- Yeah, and then, \"Whoa.\"- Oh, yeah, and then, and I'msure if your depth perceptionfor whatever reason is screwed up,if you're drunk or some other reason,and then you think you'reputting your hand on the table,and you miss it, I mean it'sthe same kind of situation.- Yeah.- But there's a-- Which is why you needto be able to predict forwardjust a little bit, and sothat's where this model-predictivecontrol stuff comes in.Predict forward what youthink's gonna happen,and if that does happen,you're in good shape.If something else happens,you better start predicting again.- So, like, regenerate a plan.(Robert laughs)- Yeah.- I mean, that also requiresa very fast feedback loopof updating what your prediction,how it matches to the actual real world.- Yeah, those thingshave to run pretty quickly.- What's the challenge ofrunning things pretty quickly,1,000 hertz, of actingand sensing quickly?- You know, there's a fewdifferent layers of that.At the lowest level, youlike to run things typicallyat around 1,000 hertz,which means that, you know,at each joint of the robot,you're measuring positionor force and then tryingto control your actuator,whether it's a hydraulicor electric motor tryingto control the force comingout of that actuator.And you wanna do that really fast,something like 1,000 hertz,and that means you can't havetoo much calculationgoing on at that joint.But that's pretty manageable these days,and it's fairly common.And then, there's another layerthat you're probablycalculating, you know,maybe at 100 hertz, maybe 10 times slower,which is now starting to lookat the overall body motionand thinking about thelarger physics of the robot.And then, there's yet anotherloop that's probably happeninga little bit slower, whichis where you start to bring,you know, your perception in, your vision,and things like that, andso you need to run allof these loops sort of simultaneously.You do have to manage your computer timeso that you can squeeze inall the calculations you needin real time in a very consistent way.And the amount of calculationwe can do is increasingas computers get better,which means we can startto do more sophisticated calculations.I can have a more complex modeldoing my forward prediction,and that might allow me todo even better predictionsas I get better and better.And it used to be, again,you know, 10 years ago,we had to have pretty simplemodels that we were running,you know, at those fast rates'cause the computers weren'tas capable about calculating forwardwith a sophisticated model,but as computation gets better,we can do more of that.- What about the actual pipelineof software engineering,how easy it is to keep updating Atlas,like, do continuous development on it?So, how many computers are on there?Is there a nice pipeline?- It's an important part ofbuilding a team around it,which means, you know, you needto also have software tools,simulation tools, you know,so we have always made strong useof physics-basedsimulation tools to do someof this calculation, basicallytest it in simulationbefore you put it on the robot.But you also want the samecode that you're runningin simulation to be thesame code you're runningon the hardware, and soeven getting to the pointwhere it was the same codegoing from one to the other,we probably didn't really get that workinguntil, you know, several years ago.But you know, that wasa bit of a milestone.And so, you wanna certainlywork these pipelinesso that you can makeit as easy as possibleand have a bunch of peopleworking in parallel.You know, we only have, youknow, four of the Atlas robots,the modern Atlas robots atthe company, and you know,we probably have, youknow, 40 developers thereall trying to gain access to it.And so, you need to share resourcesand use some of the software pipeline.- Well, that's a reallyexciting step to be able to runthe exact same code in simulationas on the actual robot.How hard is it to doa realistic simulation,physics-based simulationof Atlas such that,I mean, the dream is like,if it works in simulation,it works perfectly in reality.(Robert laughs)How hard is it to sort of keepworking on closing that gap?- The root of some of ourphysics-based simulation toolsreally started at MIT, and we builtsome good physics-basedmodeling tools there.The early days of thecompany, we were tryingto develop those toolsas a commercial product,so we continued to develop them.It wasn't a particularlysuccessful commercial product,but we ended upwith some nice physics-basedsimulation toolsso that, when we starteddoing legged robotics again,we had a really nice tool to work with.And the things we paidattention to were thingsthat weren't necessarily handled very wellin the commercial tools youcould buy off the shelf,like interaction with theworld, like foot-ground contact.And so, trying to modelthose contact events wellin a way that captured the important partsof the interaction was areally important elementto get right and to also do in a waythat was computationallyfeasible and could run fast'cause if your simulationruns too slow, you know,then your developers aresitting around waiting for stuffto run and compile, and so it's alwaysabout efficient, fast operation as well.So, that's been a big part of it.You know, I think developingthose tools in parallelto the development ofthe platform and tryingto scale them has reallybeen essential, I'd say,to us being able toassemble a team of peoplethat could do this.- Yeah, how to simulate contact, period,so foot-ground contact butsort of for manipulationbecause don't you want tomodel all kinds of surfaces?- Yeah.So, it will be even morecomplex with manipulation'cause there's a lotmore going on. (laughs)- Yeah.- You know.And you need to capture, I don't know,things slipping and moving,you know, in your hand.It's a level of complexitythat I think goesabove foot-ground contactwhen you really start doingdextrous manipulation.So, there's challenges ahead still.- So, how far are we awayfrom me being able to walkwith Atlas in the sand along the beach(Robert laughs)and us both drinking a beer?(Robert laughing)- Well, I-- Sip it out of a can, out of a can.- Maybe Atlas could spill his beer'cause he's got nowhereto put it. (laughing)Atlas could walk on the sand.- So, can it?- Yeah, yeah.Yeah, I mean, you know,have we really had himout on the beach?You know, we take them outside often,you know, rocks, hills,that sort of thing,even just around our lab in Waltham.We probably haven't beenon the sand, but I'm-- So, soft surfaces, normally?- I don't doubtthat we could deal with it.We might have to spenda little bit of timeto sort of make that work.We had to take BigDogto Thailand years ago,and we did this great videoof the robot walking in the sand,walking into the ocean upto, I don't know, its bellyor something like that andthen turning around and walkingout all while playing-- Oh, that's-- some cool beach music.- Yeah.- Great show,but then, you know, we didn'treally clean the robot off,and the saltwater was reallyhard on it, so you know,we put it in a box, shipped it back.By the time it came back,we had some problems(laughing) with corrosion.- It's the salt water.It's not like-- Salt tough (laughs).- It's not, like, sand gettinginto the components orsomething like this.- Yeah, yeah.- But I'm sure,if this is a big priority,you can make it like-- Right.- waterproof itor something.- Right, right.That just wasn't our goal at the time.- Well, it's a personal goal of mineto walk along,(Robert laughs)walk along the beach, butit's a human problem, too.You get sand everywhere.It's just a giant mess.(Robert laughs)So, soft surfaces are okay.So, I mean, can we just lingeron the robotics challenge?There's a pile of, like,rubble they had to walk over.How difficult is that task?- In the early days of developing BigDog,the loose rock was the epitomeof the hard walking surfacebecause you stepped down,and you had these littlepoint feet on the robot,and the rock can roll,and you have to dealwith that last-minute, you know, changein your foot placement.- Yeah, so you step on thething, and that thing respondsto you stepping on it.- Yeah, and it moves whereyour point of support is.And so, that became kindathe essence of the test.And so, that was thebeginning of us startingto build rock piles in our parking lots,and we would actuallybuild boxes full of rocksand bring 'em into thelab, and then we would havethe robots walking acrossthese boxes of rocksbecause that became the essential test.- So, you mentioned BigDog.Can we maybe take a strollthrough the history of Boston Dynamics?So, what and who is BigDog?By the way, is who,(Robert laughs)do you try not toanthropomorphize the robots?Do you try to remember that they're,this is like the division I have'cause, for me, it's impossible.For me, there's a magic tothe being that is a robot.It is not human, but it is,the same magic that a livingbeing has when it movesabout the world is there in the robot.So, I don't know what question I'm asking,but should I say what or who I guess?Who is BigDog?What is BigDog?(Robert laughs)- Well, I'll say toaddress the meta question,we don't try to draw hardlines around it being an it,or a him, or a her.It's okay, right?I think part of the magic ofthese kinds of machines isby nature of their organicmovement, of their dynamics,we tend to want to identify with them.We tend to look at them andsort of attribute maybe feelingto that because we've only seen thingsthat move like this that were alive.And so, this is an opportunity.It means that you couldhave feelings for a machine,and you know, people havefeelings for their cars.You know, they get attractedto 'em, attached to them.So, that's inherently,could be a good thingas long as we managewhat that interaction is.So, we don't put strongboundaries around thisand ultimately think it's a benefit,but it's also, can be a bit of a cursebecause I think peoplelook at these machines,and they attribute a level of intelligencethat the machines don't have.Why?Because, again, they'veseen things move like thisthat we're living beings,which are intelligent,and so they wanna attributeintelligence to the robotsthat isn't appropriate yet,even though they movelike an intelligent being.- But you try to acknowledgethat the anthropomorphizationis there and tryto, first of all,acknowledge that it's there.- And have a little fun with it.- And have little fun.- You know,our most recent video,it's just kind of fun,you know, to look at the robot.We started off the videowith Atlas kind of lookingaround for where the bag of tools was'cause the guy up on the scaffolding says,\"Send me some tools.\"Atlas has to kinda lookaround and see where they are.And there's a littlepersonality there that is fun.It's entertaining.It makes our jobs interestingand I think in the longrun can enhance interactionbetween humans and robots in a waythat isn't available to machinesthat don't move that way.- This is something to mepersonally is very interesting.I happen to have a lot of legged robots.(both laughing)I hope to have a lot ofSpots in my possession.I'm interested in celebrating roboticsand celebrating companies, companiesthat do incredible stufflike Boston Dynamics.You know, I'm a little crazy,and you say you don't want to,you want to align, youwanna help the company'cause I ultimately want a companylike Boston Dynamics to succeed.And part of that we'lltalk about, you know,success kind of requires making money.And so, the kinda stuffI'm particularly interestedin may not be the thing thatmakes money in the short term.I can make an argumentthat will in the long term.But the kinda stuff I've beenplaying with is a robust wayof having the quadrupeds, therobot dogs communicate emotionwith their body movement-- Hmm.- the same kinda stuffyou do with a dog-- Yeah.- but not hard coded,but in a robust way-- Mm-hmm.- and be able to communicateexcitement, or fear-- Mm-hmm.- boredom, all these kinds of stuff.And I think as a base layerof function, of behaviorto add on top of a robot, I think that'sa really powerful wayto make the robot more usable for humans,for whatever application.- I think it's gonna bereally important, and it'sa thing we're beginningto pay attention to.A differentiator for thecompany has always beenwe really want the robot to work.We want it to be useful.Making it work at first meantthe legged locomotion really works.It can really get around,and it doesn't fall down.But beyond that, now itneeds to be a useful tool.And our customers are, forexample, factory owners,people who are running aprocess-manufacturing facility.And the robot needs to be able to getthrough this complexfacility in a reliable way,you know, taking measurements.We need for people whoare operating those robotsto understand what the robots are doing.If the robot needs helpor, you know, is in troubleor something, it needsto be able to communicateand a physical indication of some sortso that a person looksat the robot and goes,\"Oh, I know what that robot's doing.That robot's going to go take measurementsof my vacuum pump withits thermal camera.\"You know, you wanna beable to indicate that,or even just the robot'sabout to turn, you know,in front of you andmaybe indicate (laughs)that it's going to turn.And so, you sort of see andcan anticipate its motion.So, this kind of communication is goingto become more and more important.It wasn't sort of ourstarting point, you know,but now the robots arereally out in the world,and you know, we have about 1,000 of 'emout with customers right now.This layer of physical indication,I think, is gonna becomemore and more important.- We'll talk about where it goes'cause there's a lot ofinteresting possibilities.But if you can return back tothe origins of Boston Dynamicswith the more research, the R&D side,before we talk about howto build robots at scale.So, BigDog.What's-- So-- Who's BigDog?- So, the company started in 1992,and in probably 2003,I believe, is when wetook a contract from,so, basically, 10 years, 11years we weren't doing robotics.We did a little bit of robotics with Sony.They had Aibo.They had their Aibo robot.We were developing some software for that.That kinda got us a little bit involvedwith robotics again.Then, there was this opportunityto do a DARPA contractwhere they wanted to build a robot dog.And we won a contract to build that.And so, that was the genesis of BigDog,and it was a quadruped,and it was the first timewe built a robot thathad everything on board.You could actually take the robotout into the wild and operate it.So, it had an onboard power plan.It had onboard computers.It had hydraulic actuatorsthat needed to be cooled.So, we had cooling systems built in,everything integrated into the robot.And that was a pretty rough start, right?It was 10 years that wewere not a robotics company.We were a simulationcompany, and then we hadto build a robot in about a year.So, that was a little bitof a rough transition.(Lex laughs)(Robert laughs)- I mean, can you justcomment on the roughnessof that transition?'Cause BigDog, I mean,this is this bigquadruped, four legs robot.- We built a few differentversions of them,but the very earliest ones, youknow, didn't work very well,(laughs) and we would take 'emout, and it was hard to get,you know, a go-kart enginedriving a hydraulic-- Oh, is that what it was?(Robert laughs)I was-- And you know,having that all work while trying to get,you know, the robot to stabilize itself,and so-- So.what was the power plan?What was the engine?It seemed like my vaguerecollection, (laughs)I don't know.It felt very loud, and aggressive,and kind of thrown togetheris what it kind of-- Oh, it absolutely was, right?We weren't trying to designthe best robot hardwareat the time, and we wanted tobuy an off-the-shelf engine.And so, many of the earlyversions of BigDog had literallygo-kart engines or something like that.Usually, it-- It was gas powered?- Yeah, a gas-powered two-stroke engine.(Lex laughs)And the reason why it was two stroke istwo-stroke engines are lighter weight,and we generally didn'tput mufflers on them'cause we're trying to save the weight,and we didn't care about the noise.(laughing) And some of thesethings were horribly loud,but we're trying to manageweight because managing weightin a legged robot is always importantbecause it has to carry everything.- That said, that thing was big-- Well-- what I've seenthe videos of.- Yeah.I mean, the earlyversions, you know, stoodabout, I don't know,belly high, chest high.You know, they probablyweighed maybe a coupleof hundred pounds, butyou know, over the courseof probably five years, wewere able to get that robotto really manage a remarkablelevel of rough terrain.So, you know, we started outwith just walking on the flat,and then we started walkingon rocks, and then inclines,and then mud, and then slippery mud.And you know, by the end ofthat program, we were convincedthat legged locomotion ina robot could actually work'cause you know, going intoit, we didn't know that.We had built quadrupeds at MIT,but they used a giant hydraulicpump, you know, in the lab.They used a giant computerthat was in the lab.They were always tethered to the lab.This was the first timesomething that was sortof self-contained, you know,walked around in the worldand balanced, and the purposewas to prove to ourselfthat the legged locomotioncould really work.And so, BigDog reallycut that open for us.And it was the beginning of what becamea whole series of robots.So, once we showed toDARPA that you could makea legged robot that could work,there was a period at DARPAwhere robotics got reallyhot, and there was lotsof different programs,and you know, we were ableto build other robots.We built other quadrupedslike LS3 designedto carry heavy loads.We built Cheetah, whichwas designed to explore,what are the limits tohow fast you can run?You know, we began tobuild sort of a portfolioof machines and software that let us buildnot just one robot, buta whole family of robots.- So, push the limits inall kinds of directionsin terms-- Yeah,and to discover those principles.You know, you asked earlierabout the art and scienceof legged locomotion.We were able to developprinciples of legged locomotionso that we knew how to builda small legged robot or a big one.So, leg length, youknow, was now a parameterthat we could play with.Payload was a parameterwe could play with.So, we built the LS3, whichwas an 800-pound robotdesigned to carry a 400-pound payload.And we learned the design rules,basically developed the design rules.How do you scale different robot systemsto, you know, their terrain,to their walking speed,to their payload?- So, when was Spot born?- Around 2012 or so,so, again, almost 10 yearsinto sort of a run with DARPAwhere we built a bunchof different quadrupeds.We had sort of a different threadwhere we started building humanoids.We saw that probably an end was comingwhere the government wasgonna kind of back offfrom a lot of robotics investment.And in order to maintainprogress, we just deducedthat, \"Well, we probablyneed to sell ourselvesto somebody who wants tocontinue to invest in this area,\"and that was Google.And so, at Google, we wouldmeet regularly with Larry Page,and Larry just startedasking us, you know,\"What's your product gonna be?\"And you know, the logical thing, the thingthat we had the mosthistory with that we wantedto continue developing was a quadruped.But we knew it needed to be smaller.We knew it couldn't have a gas engine.We thought it probably couldn'tbe hydraulically actuated.So, that began the process ofexploring if we could migrateto a smaller, electrically actuated robot.And that was really the genesis of Spot.- So, not a gas engine, andthe actuators are electric.- Yes.- So, can you maybecomment on what it's likeat Google working with Larry Page,having those meetings, and thinkingof what will a robot look likethat could be built at scale,like, starting to think about a product?- Larry always liked the toothbrush test.He wanted products thatyou used every day.What they really wanted was, you know,a consumer-level product,something that would work in your house.We didn't think that wasthe right next thing to do,because to be a consumer-level product,cost is gonna be very important.Probably needed to costa few thousand dollars.And we were building these machinesthat cost hundreds ofthousands of dollars,maybe a million dollars to build.And of course, we wereonly building, like, two,but we didn't see how to get all the wayto this consumer-level product-- In a short amount of time.- In a short amount of time.And he suggested that we makethe robots really inexpensive,and part of our philosophy has always beenbuild the best hardware you can.Make the machine operatewell so that you're tryingto solve, you know,discover the hard problemthat you don't know about.Don't make it harderby building a crappy machine, basically.Build the best machine you can.There's plenty of hard problems to solvethat are gonna have to dowith, you know, underactuatedsystems and balance.And so, we wanted to buildthese high-quality machines still,and we thought that was important for usto continue learning about really what wasthe important parts that make robots work.And so, there was a little bitof a philosophical difference there.And so, ultimately that'swhy we're building robotsfor the industrial sector nowbecause the industry canafford a more expensive machinebecause, you know, theirproductivity dependson keeping their factory going.And so, if Spot costs, youknow, $100,000 or more,that's not such a big expense to them,whereas at the consumer level,no one's gonna buy a robot like that.And I think we might eventually getto a consumer-level productthat will be that cheap,but I think the path toget in there needs to gothrough these really nice machinesso that we can then learn how to simplify.- So, what can you say toalmost the engineering challengeof bringing down cost of a robotso that, presumably, when youtry to build a robot at scale,that also comes into play whenyou're trying to make moneyon a robot even in the industrial setting?But how interesting, howchallenging of a thing is that,in particular probably newto an R&D company?(Robert laughs)- Yeah, I'm glad youbrought that last part up.The transition from an R&Dcompany to a commercial company,that's the thing youworry about, you know,'cause you've got theseengineers who love hard problems,who wanna figure outhow to make robots work.And you don't know if youhave engineers that wanna workon the quality, and reliability, and costthat is ultimately required.And indeed, you know, we havebrought on a lot of new peoplewho are inspired by those problems.But the big takeaway lesson for me iswe have good people.We have engineers whowanna solve problems,and the quality, and cost,and manufacturability isjust another kind of problem.And because they're soinvested in what we're doing,they're interested in and will go workon those problems as well.And so, I think we're managingthat transition very well.In fact, I'm really pleased that, I mean,it's a huge undertaking by the way, right?So, you know, to get reliabilityto where it needs to be,we have to have fleets of robotsthat we're just operating24/7 in our officesto go find those rarefailures and eliminate them.It's just a totallydifferent kind of activitythan the research activitywhere you get it to work,you know, the one robot you have to workin a repeatable way, (laughs) you know,at the high-stakes demo.It's just very different.But I think we're makingremarkable progress, I guess.- So, one of the coolthings, I got a chanceto visit Boston Dynamics, and I mean,one of the things that's really cool isto see a large numberof robots moving aboutbecause I think one ofthe things you noticein the research environmentat MIT, for example,I don't think anyoneever has a working robotfor a prolonged period of time.- (laughing) Exactly.- So, like, most robotsare just sitting therein a sad state of despairwaiting to be born,(Robert laughs)brought to life for abrief moment of time.I just remember there'sa Spot robot just had,like, a cowboy hat on andwas just walking randomlyfor whatever reason.I don't even know, butthere's a kind of a senseof sentience to it because it doesn't seemlike anybody was(laughing) supervising it.- Well-- It was just doingits thing.- I'm gonna stopway short of the sentience.- Sure.- It is the casethat, if you come to ouroffice, you know, todayand walk around the hallways,you're gonna see a dozen robotsjust kind of walking around-- Yes.- all the time.And that's really areliability test for us.So, we have these robots programmedto do autonomous missions, getup off their charging dock,walk around the building, collect dataat a few different places,and go sit back down.And we want that to bea very reliable process'cause that's what somebodywho's running a brewery,a factory, that's whatthey need the robot to do,and so we have to dog-food our own robot.We have to test it in that way.And so, on a weekly basis, wehave robots that are accruingsomething like 1,500 or maybe2,000 kilometers of walkingand you know, over 1,000hours of operation every week.And that's something thatI don't think anybody elsein the world can do 'cause,A, you have to have a fleetof robots to just accrue thatmuch information. (laughing)You have to be willing todedicate it to that test.But that's essential.- That's how youget the reliability.- That's how you get it.- What about some of the cost cuttingfrom the manufacturer's side?What have you learned fromthe manufacturer's sideof the transition from R&D to-- And we're still learning a lot there.We're learning how to cast partsinstead of mill it all outof, you know, billet aluminum.We're learning how toget plastic molded parts,and we're learning abouthow to control that process(laughs) so that you can buildthe same robot twice in a row.There's a lot to learn there.And we're only partwaythrough that process.We've set up a manufacturingfacility in Waltham.It's about a mile from our headquarters,and we're doing final assembly and testsof both Spots and Stretches,you know, at that factory.And it's hard because, to behonest, we're still iteratingon the design of the robot.As we find failures fromthese reliability tests,we need to go engineerchanges, and those changes needto now be propagated tothe manufacturing line.And that's a hard process,especially when you wannamove as fast as we do.And that's been challenging.You know, the folks whoare working supply chainwho are trying to get thecheapest parts for us, kindof requires that you buy alot of 'em to make 'em cheap,and then we go change thedesign from underneath 'em,and they're like, \"What are you doing?\"And so, you know, gettingeverybody on the same page herethat, yep, we still need to move fast,but we also need to try tofigure out how to reduce cost,that's one of the challengesof this migration we're going through.- And over the past few years,challenges to the supply chain, I mean,I imagine you've been a partof a bunch of stressful meetings.- Yeah, things got moreexpensive and harder to get,and yeah, so it's all been a challenge.- Is there still room for simplification?- Oh, yeah, much more,and you know, these arereally just the firstgeneration of these machines.We're already thinking aboutwhat the next generationof Spot's gonna look like.Spot was built as aplatform, so you could putalmost any sensor on it.You know, we provided data communications,mechanical connections, power connections.But for example, in theapplications that we're excitedabout where you'remonitoring these factoriesfor their health, there'sprobably a simpler machinethat we could build that'sreally focused on that use case.And that's the differencebetween the general-purposemachine or the platformversus the purpose-built machine.And so, even though even in the factorywe'd still like the robot todo lots of different tasks,if we really knew on day onethat we're gonna be operatingin a factory with thesethree sensors in it,we would have it allintegrated in a packagethat would be easier, lessexpensive, and more reliable.So, we're contemplatingbuilding, you know,a next generation of that machine.- So, we should mention that,so Spot for people whosomehow are not familiar, isa yellow, robotic dogand has been featuredin many dance videos.It also has gained an arm.So, what can you say aboutthe arm that Spot has,about the challenges of this design,and the manufacturer of it?- We think the future of mobile robots ismobile manipulation.You know, in the past 10 years,it was getting mobility to work,getting the legged locomotion to work.If you ask, what's the hardproblem in the next 10 years,it's getting a mobile robotto do useful manipulation for you.And so, we wanted Spot to have an armto experiment with those problems.And the arm is almost ascomplex as the robot itself,you know, and it's an attachable payload.It has, you know, several motors,and actuators, and sensors.It has a camera in the end of its hand,so you know, you cansort of see something,and the robot will controlthe motion of its handto go pick it up autonomously.So, in the same way therobot walks and balances,managing its own footplacement to stay balanced,we want manipulation tobe mostly autonomous,where the robot, you indicate,\"Okay, go grab that bottle,\"and then the robot will justgo do it using the camerain its hand and then sortof closing in on the grasp.But it's a whole nother complex roboton top of a complex legged robot.And of course, we made thehand look a little like a head,(laughs) you know,because again, we want itto be sort of identifiable.In the last year, a lot ofour sales have been peoplewho already have a robot now buying an armto add to that robot.- Oh, interesting.And so, the arm is for sale?- Oh, yeah, oh, yeah.It's an option.- What's the interfacelike to work with the arm?I could just ask that question in generalabout robots from Boston Dynamics.Is it designed to be easilyand efficiently operatedremotely by a human being?Or, is there also the capabilityto push towards autonomy?- We want both.In the next version of thesoftware that we release,which will be version 3.3,we're gonna offer the ability,if you have a autonomousmission for the robot,we're gonna include theoption that it can gothrough a door, which meansit's gonna have to have an arm,and it's gonna have to usethat arm to open the door.And so, that'll be anautonomous manipulation taskthat you can programeasily with the robot-- Oh.- strictly through, you know,we have a tablet interface,and so on the tablet, you know,you sort of see the view that Spot sees.You say, \"There's the door handle.You know, the hinges are onthe left, and it opens in.The rest is up to you.Take care of it.\"- Oh, wow.So, it just takes care of everything?- Yeah.So, and for a task like opening doors,you can automate most of that.And we've automated a few other tasks.We had a customer who hada high-powered breakerswitch, essentially.It's an electric utility,Ontario Power Generation.And when they're gonnadisconnect, you know,their power supply, right,that could be a gas generator,could be a nuclear powerplant, you know, from the grid,you have to disconnectthis breaker switch.Well, as you can imagine,there's, you know,hundreds or thousands of ampsand volts (laughing) involvedin this breaker switch.And it's a dangerous event'cause occasionally you'll getwhat's called an arc flash.As you just do this disconnect, the power,the sparks jump across,and people die doing this.And so, Ontario Power Generationused our Spot and the armthrough the interface tooperate this disconnect-- That's great.- in an interactive way.And they showed it to us, andwe were so excited about itand said, \"You know, I betwe can automate that task.\"And so, we got some examplesof that breaker switch,and I believe in the nextgeneration of the softwarenow we're gonna deliver backto Ontario Power Generation,they're gonna be ableto just point the robotat that breaker.They'll indicate, \"That's the switch.\"There's sort of twoactions you have to do.You have to flip up thislittle cover, press a button,then get a ratchet,stick it into a socket,and literally unscrewthis giant breaker switch.So, there's a bunch of different tasks,and we basically automatedthem so that the human says,\"Okay, there's the switch.Go do that part.That right there is the socketwhere you're gonna put your tool,and you're gonna open it up.\"And so you can remotelysort of indicate thison the tablet, and thenthe robot just doeseverything in between.- And it does everything,all the coordinated movementof all the different actuatorsthat includes the bodyand the arm.- Yeah, maintains its balance.It walks itself, you know, into positionso it's within reach, andthe arm is in a positionwhere it can do the whole task.So, it manages the whole body.- So, how does one becomea big enough customerto request features?'Cause I personally want arobot that gets me a beer.(Robert laughs)I mean, that has to be,like, one of the most,I suppose, in the industrial setting,that's a non-alcoholic beverageof picking up objects andbringing the objects to you.- We love working with customerswho have challenging problems like thisand this one in particular because we feltlike what they were doing,A, it was a safety feature.B, we saw that the robot could do it'cause they teleoperatedit the first time.Probably took 'em an hour todo it the first time, right?But the robot was clearlycapable, and we thought,\"Oh, this is a greatproblem for us to work onto figure out how to automatea manipulation task.\"And so, we took it on not becausewe were gonna make a bunchof money from it in sellingthe robot back to thembut because it motivatedus to go solve what we sawas the next logical step.But many of our customers, in fact,our bigger customers, typically oneswho are gonna run a utility, or a factory,or something like that, wetake that kind of directionfrom them, especiallyif they're gonna buy 10,or 20, or 30 robots, and they say,\"I really need it to dothis,\" well, that's exactlythe right kind of problemthat we wanna be working on.- Mm-hmm.- Yeah, and so-- Note to self, \"Buy 10 Spots,(Robert laughs)and aggressively pushfor beer manipulation.\"(Robert laughs)I think it's fair to sayit's notoriously difficultto make a lot of moneyas a robotics company.How can you make moneyas a robotics company?Can you speak to that?It seems that a lot ofrobotics companies fail.It's difficult to build robots.It's difficult to buildrobots at a low enough costwhere customers, even inthe industrial setting, wantto purchase them, and it'sdifficult to build robotsthat are useful, sufficiently useful.- Yeah.- So, what can you speak to?And Boston Dynamics has beensuccessful for many yearsof finding a way to make money.- Well, in the earlydays, of course, you know,the money we made was fromdoing contract R&D work,and we made money, but youknow, we weren't growing,and we weren't selling a product.And then, we went through several ownerswho, you know, had a visionof not only developingadvanced technology, buteventually developing products.And so, both, you know,Google, and SoftBank,and now Hyundai, you know, hadthat vision and were willingto, you know, provide that investment.Now, our discipline is that weneed to go find applicationsthat are broad enough thatyou could imagine sellingthousands of robotsbecause it doesn't workif you don't sell thousands ortens of thousands of robots.If you only sell hundreds,you will commercially fail.And that's where mostof the small robot companies have died.And that's a challenge because, you know,A, you need to field the robots.They need to start to becomereliable, and as we've said,that takes time andinvestment to get there.And so, it really does takevisionary investment to get there.But we believe that weare going to make moneyin this industrial monitoring spacebecause, you know, if a chip fab,if the line goes downbecause a vacuum pump failed someplace,that can be a very expensive process.It can be a million dollarsa day in lost production,maybe you have to throw away someof the product along theway, and so the robot,if you can prevent thatby inspecting thefactory every single day,maybe every hour if you have to,there's a real return on investment there.But there needs to be acritical mass of this task.And we're focusing on a fewthat we believe are ubiquitousin the industrial production environment.And that's using a thermalcamera to keep thingsfrom overheating, using an acoustic imagerto find compressed airleaks, using visual camerasto read gauges, measuring vibration.These are standard things that you doto prevent unintendedshutdown of a factory.And this takes place in a beer factory.We're working with AB InBev.It takes place in chip fabs.You know, we're workingwith GlobalFoundries.It takes place in electric utilitiesand nuclear power plants.And so, the same robot can be appliedin all of these industries.And as I said, we haveabout, actually it's 1,100 Spots out now.To really get, you know,profitability, we need to beat 1,000 a year, maybe1,500 a year, you know,for that sort of part of the business.So, it still needs to grow,but we're on a good path.So, I think that's totally achievable.- So, the applicationshould require crossingthat 1,000-robot barrier.- It really should, yeah.I wanna mention, you know,our second robot, Stretch.- Yeah, tell me about Stretch.What's Stretch?Who is Stretch?- Stretch started differently than Spot.You know, Spot we builtbecause we had decadesof experience building quadrupeds.We had it in our blood.We had to build a quadruped product,but we had to go figure outwhat the application was,and we actually discovered thisfactory-patrol application,basically preventative maintenanceby seeing what our customers did with it.Stretch was very different.We started knowingthat there was warehousesall over the world.There's shipping containersmoving all around the world fullof boxes that are mostlybeing moved by hand.By some estimates, we thinkthere's a trillion boxes,(laughs) cardboard boxes shippedaround the world each year.And a lot of it's done manually.It became clear early onthat there was an opportunityfor a mobile robot inhere to move boxes around.And the commercial experiencehas been very differentbetween Stretch and with Spot.As soon as we started talkingto people, potential customersabout what Stretch was gonna be used for,they immediately started saying,\"Oh, I'll buy that robot.You know, in fact, I'mgonna put in an orderfor 20 right now.\"We just started shippingthe robot in Januaryafter, you know, severalyears of development.- Of this year?- Of this year.So, our first deliveries ofStretch to customers wereDHL and Maersk in January.We're delivering to Gapright now, and we haveabout seven or eight other customers,all who've alreadyagreed in advance to buybetween 10 and 20 robots,and so we've already got commitmentsfor, you know, a couplehundred of these robots.This one's gonna go, right?It's so obvious that there's a need,and we're not just gonna unload trucks.We're gonna do any box-movingtask in the warehouse.And so, it too will bea multipurpose robot,and we'll eventually haveit doing palletizing,or depalletizing, or loadingtrucks, or unloading trucks.There's definitely thousands of robots.There's probably tensof thousands of robotsof this in the future, soit's gonna be profitable.- Can you describe whatStretch looks like?- It looks like a big, strongrobot arm on a mobile base.The base is about the size of a pallet,and we wanted it to bethe size of a palletbecause that's what livesin warehouses, right,pallets of goods sitting everywhere,so it needed to be ableto fit in that space.- It's not a legged robot.- It's not a legged robot.So, it was our first,it was actually a bitof a commitment from us,a challenge for us to builda non-balancing robot.(Lex laughs)- To do the much easierproblem but to do it well.- Well, because, you know,it wasn't gonna have this balance problem.And in fact, the very first versionof the logistics robot webuilt was a balancing robot,and that's called Handle.And there's-- That thing was epic.- Oh, it's a beautiful machine.- It's an incredible machine.(Robert laughs)(Lex laughs)I mean, it looks epic.It looks like, I mean,out of a sci-fi movie of some sorts.I mean, can you actually just lingeron, like, the design of that thing?'Cause that's another leapinto something you probably haven't done.It's a different kind of balancing.- Yeah, so let me-- It's wild.- I love talking about thehistory of how Handle came about(Lex laughs)because it connects allof our robots, actually.So, I'm gonna start with Atlas.When we had Atlasgetting fairly far along,we wanted to understand,I was telling you earlier,the challenge of the human form isthat you have this mass up high,and balancing that inertia,that mass up high is itsown unique challenge.And so, we started trying to get Atlasto balance standing on one foot,like on a balance beamusing its arms like this,and you know, you can do this, I'm sure.I can do this, right?Like, if you're walking a tightrope,how do you do that balance?So, that's sort of, you know,controlling the inertia,controlling the momentum of the robot.We were starting tofigure that out on Atlas.And so, our first concept of Handle,which was a robot that wasgonna be on two wheels,so it had to balance, but itwas gonna have a big, long armso it could reach a boxat the top of a truck,and it needed yet anothercounterbalance, a big tail,to help it balance whileit was using its arm.So, the reason why thisrobot sort of looks epic,some people said it looked like an ostrichor maybe, you know, anostrich moving around, wasthe wheels, the leg.It has legs, so it can extend its legs.So, it's wheels on legs.We always wanted to build wheels on legs.It had a tail, and it had this arm,and they're all movingsimultaneously and in coordinationto maintain balance because we had figuredout the mathematics ofdoing this momentum control,how to maintain that balance.And so, part of the reasonwhy we built this two-legged robot waswe had figured this thing out.We wanted to see it inthis kind of machine,and we thought maybe thiskind of machine would be goodin a warehouse, and so we built it.And it's a beautiful machine.It moves in a graceful waylike nothing else we've built.But it wasn't the right machinefor a logistics application.We decided it was too slowand couldn't pick boxesfast enough, basically.- Oh.- And it-- Do it beautifullywith elegance.- It did beautifully,but it just wasn't efficient enough.- Aw.- So, we let it go.- Yeah.- But I think we'll come backto that machine eventually.- The fact that it's possible,the fact that you showedthat you could do so manythings at the same timein coordination and so beautifully,there's something there.- Yeah.- That was a demonstrationof what is possible.- Basically, we made a hard decision,and this was really kind of ahard-nosed business decision.It indicated us not doingit just for the beautyof the mathematics or the curiosity,but no, we actuallyneed to build a businessthat can make money in the long run.And so, we ended up building Stretch,which has a big, heavy basewith a giant battery in the baseof it that allows itto run for two shifts,16 hours worth of operation.And that big battery sort ofhelps it stay balanced, right?So, it can move a 50-poundbox around with its armand not tip over it.It's omnidirectional,it can move in any direction,and it has a nice suspensionbuilt into it so it candeal with, you know, gapsor things on the floor and roll over it.But it's not a balancing robot.It's a mobile robot arm thatcan work to carry, or pick,or place a box up to 50 poundsanywhere in the warehouse.- Take a box from pointA to point B anywhere.- Yeah, palletize, depalletize.We're starting with unloading trucksbecause there's so manytrucks and containerswhere goods are shipped,and it's a brutal job.You know, in the summer,it can be 120 degreesinside that container.People don't wanna do that job,and it's backbreaking labor, right?Again, these can be up to 50-pound boxes.And so, we feel like thisis a productivity enhancer,and for the people who used todo that job unloading trucks,they're actually operating the robot now.And so, by building robotsthat are easy to control,and it doesn't take anadvanced degree to manage,you can become a robot operator.And so, as we've introduced these robotsto both DHL, and Maersk, andGap, the warehouse workerswho were doing that manual labor arenow the robot operators,and so we see thisas ultimately a benefit to them as well.- Can you say how much Stretch costs?- Not yet, but I willsay that, when we engagewith our customers, they'llbe able to see a returnon investment in typically two years.- Okay, so that's somethingthat you're constantly thinkingabout, how-- Yeah.- And I suppose youhave to do the same kindof thinking with Spot.So-- Yes.- it seems like withStretch the application is,like, directly obvious.- Yeah, it's a slam dunk.- Yeah, and so you have alittle more flexibility.- Well, I think we know the target.We know what we're going after.- Yeah.- And with Spot, it tookus a while to figureout what we were going after.- Well, let me return to that questionabout maybe theconversation you were havinga while ago with Larry Page, maybe lookingto the longer future ofsocial robotics, of using Spotto connect with humanbeings perhaps in the home.Do you see a future thereif we were to sort of hypothesizeor dream about a futurewhere Spot-like robotsare in the home as pets,a social robot?- We definitely thinkabout it, and we would like to get there.We think the pathway togetting there is, you know,likely through theseindustrial applicationsand then mass manufacturing, you know.Let's figure out how to build the robots,how to make the softwareso that they can really doa broad set of skills.That's going to take realinvestment to get there.Performance first, right?A principle of the company has always beenreally make the robots do useful stuff.And so, you know, thesocial robot companiesthat try to start someplace elseby just making a cute interaction,mostly they haven't survived.And so, we think the utilityreally needs to come first,and that means you have to solve someof these hard problems.And so, to get there, we'regonna go through the designand software development in industrial,and then that's eventuallygonna let you reach a scalethat could then beaddressed to a commercial,a consumer-level market, andso, yeah, maybe we'll be ableto build a smaller Spot with an armthat could really goget your beer for you.- Mm-hmm.- But there's things weneed to figure out still,how to safely, really safely,and if you're gonna beinteracting with children,you better be safe. (laughs)And right now, we count on a little bitof standoff distancebetween the robot and peopleso that you don't pinch afinger, you know, in the robot.So, you've got a lot ofthings you need to go solvebefore you jump to thatconsumer-level product.- Well, there's a kindof trade off in safetybecause it feels like, inthe home, you can fall.Like, you don't have to be as good.Like, you're allowed tofail in different ways,in more ways as long asit's safe for the humans.So, it just feels like aneasier problem to solve'cause it feels like, in the factory,you're not allowed to fail.- That may be true, butI also think the varietyof things a consumer-levelrobot would be expectedto do will also be quite broad.- Yeah.- And they're gonna want to get the beerand know the difference betweenthe beer and a Coca-Colaor my snack.You know, they're all gonnawant you to clean up the dishes,you know, from the tablewithout breaking 'em. (laughs)Those are pretty complex tasks,and so there's stillwork to be done there.- So, to push back on that,here's what applicationI think that'll be very interesting.I think the applicationof being a pet, a friend,so, like, no tasks.Just be cute, not cute, not cute.A dog is more than just cute.A dog is a friend, is a companion.There's something about justhaving interacted with them.And maybe 'cause I'm hanging out alonewith robot dogs a little too much,but, like, there's a connection there.And it feels like that connectionshould not be disregarded.You-- No.It should not be disregarded.Robots that can somehow communicatethrough their physical gesturesyou're gonna be moreattached to in the long run.Do you remember Aibo-- Mm-hmm.- the Sony Aibo?- Yep.- They soldover 100,000 of those, maybe 150,000,you know, what probably wasn't considereda successful product for them.They suspended that eventually,and then they brought it back.Sony brought it back,and people definitely,you know, treated thisas a pet, as a companion.And I think that will come around again.Will you get away withouthaving any other utility?Maybe in a world where we can really talkto our simple little petbecause, you know, ChatGPTor some other generativeAI has made it possiblefor you to really talk in whatseems like a meaningful way.Maybe that'll open thesocial robot up again.That's probably not apath we're gonna go downbecause, again, we're so focusedon performance and utility.We can add those other things also,but we really wanna startfrom that foundation of utility, I think.- Yeah.But I also wanna predictthat you're wrong on that,which is that the very path you're taking,which is creating a great robot platform,will very easily take a leap to addinga ChatGPT-like capability, maybe GPT 5.And there's just so manyopen-source alternativesthat you could just plopdown on top of Spot.And because you have this robust platform,and you're figuring outhow to mass-manufacture it,and how to drive the cost down,and how to make it, you know, reliable,all those kinds of things,it'll be the natural transitionto where just adding ChatGPTon top of it could-- Oh, I do thinkthat being able to verbally converseor even converse through gestures,you know, part of these learning models isthat, you know, you can nowlook at video and imageryand associate, you know, intent with that.Those will all help in the communicationbetween robots and people, for sure.And that's gonna happenobviously more quicklythan any of us were expecting. (laughs)- I mean, what else do you want from life?A friend to get you a beer(Robert laughs)and then just talk shitabout the state of the world.(Robert laughs)I mean, there's a deeploneliness within all of us.And I think a beer and a goodchat solves so much of itor takes us a long way to solvinga lot of it.- It'll be interestingto see, you know,when a generative AI can give youthat warm feeling thatyou connected, you know,and that, \"Oh, yeah, you remember me.You're my friend.You know, we have a history.\"You know, that history matters, right?- Memory of joint, like-- Memory of, yeah. (laughs)- Having witnessed,that's what friendship,that's what connection,that's what love is.In many cases, some of thedeepest friendships you have ishaving gone through adifficult time together-- Mm-hmm.- and having a shared memoryof an amazing time or a difficult timeand kind of that memorycreating this, like, foundationbased on which you can thenexperience the world together.The silly, the mundane stuffof day to day is somehow builton a foundation of having gonethrough some shit in the past.And the current systems arenot personalized in that waybut-- Right.- I think that's a technical problem,not some kind of fundamental limitation,so combine that with anembodied robot like Spot,which already has magic in its movement,I think it's a veryinteresting possibilityof where that takes us.But of course, you have tobuild that on top of a companythat's making moneywith real applications,with real customers, andwith robots that are safe,and work, and reliable,and manufactured at scale.- And I think we're in aunique position in thatbecause of, you know, ourinvestors, primarily Hyundai,but also SoftBank still owns 20% of us.They're not totally fixatedon driving us to profitabilityas soon as possible.That's not the goal.The goal really is alonger-term vision of creating,you know, what doesmobility mean in the future?How is this mobile robottechnology going to influence us,and can we shape that?And they want both.And so, we as a company aretrying to strike that balancebetween, \"Let's build abusiness that makes money.\"I've been describing thatto my own team as self-destination.If I wanna drive my own ship,we need to have a businessthat's profitable in the end.Otherwise, somebody else isgonna drive the ship for us.So, that's really important.But we're gonna retain theaspiration that we're gonna buildthe next generation oftechnology at the same time.And the real trick willbe if we can do both.- Speaking of ships, let meask you about a competitorand somebody who's become a friend.So, Elon Musk and Teslahave announced they've beenin the early days ofbuilding a humanoid robot.How does that change thelandscape of your work?So, from an outsideperspective, it seems like,well, as a fan of robotics,it just seems exciting.- Right, very exciting, right?When Elon speaks, people listen.And so, it suddenly broughta bright light onto the workthat we'd been doing, youknow, for over a decade.And I think that's only gonna help.And in fact, what we've seenis that, in addition to Tesla,we're seeing a proliferationof robotic companies arise now.- Including humanoid?- Yes.- Oh, wow.- Yeah.And interestingly, many of themas they're, you know,raising money, for example,will claim whether or not they havea former Boston Dynamics employeeon their staff as a criteria.(both laughing)- Yeah, that's true.I would do that as acompany, yeah, for sure.- Yeah, so-- Shows you're legit, yeah.- Yeah, so, you know,(Lex laughs)it has brung a tremendous validationto what we're doing and excitement.Competitive juices are flowing,you know, the whole thing.So, it's all good.- Elon has also kind of statedthat, you know,maybe he implied thatthe problem is solvablein the near term, which isa low-cost humanoid robotthat's a relativelygeneral use case robot.So, I think Elon is known forsort of setting these kindsof incredibly ambitiousgoals, maybe missing deadlinesbut actually pushing not justthe particular team he leadsbut the entire world to,like, accomplishing those.Do you see Boston Dynamics inthe near future being pushedin that kind of way, likethis excitement of competitionkinda pushing Atlas maybeto do more cool stuff,trying to drive the costof Atlas down perhaps?I mean, I guess I wannaask if there's some kindof exciting energy in Boston Dynamicsdue to this little bit of competition.- Oh, yeah, definitely.When we released our mostrecent video of Atlas, you know,I think you had seen it, the scaffoldingand throwing the box of tools aroundand then doing the flip at the end,we were trying to show the worldthat not only can we dothis parkour mobility thing,but we can pick up and move heavy thingsbecause, if you're gonna workin a manufacturing environment,that's what you gotta be able to do.And for the reasons Iexplained to you earlier,it's not trivial to do so, you know,changing the center of mass, you know,by picking up a 50-pound block, you know,for a robot that weighs 150 pounds,that's a lot to accommodate.So, we're trying to showthat we can do that,so it's totally been energizing.You know, we see thenext phase of Atlas beingmore dextrous hands that canmanipulate and grab more thingsthat we're gonna start bymoving big things aroundthat are heavy and that affect balance.And why is that?Well, really tiny dextrousthings probably are gonna be hardfor a while yet, you know.Maybe you could go build aspecial-purpose robot arm,you know, for stuffing, you know, chipsinto electronics boards,but we don't really wannareally fine work like that.I think more course workwhere you're using two handsto pick up and balance an unwieldy thingmaybe in a manufacturing environment,maybe in a construction environment,those are the things that wethink robots are gonna be ableto do with the level ofdexterity that they're gonna havein the next few years, andthat's where we're headed.And you know, Elon hasseen the same thing, right?He's talking about using the robotsin a manufacturing environment.We think there's somethingvery interesting thereabout having a two-armed robotbecause, when you have twoarms, you can transfer a thingfrom one hand to the other.You can turn it around.You know, you can reorient itin a way that you can't do itif you just have one handon it, and so there's a lotthat extra arm brings to the table.- So, I think in terms of mission,you mentioned BostonDynamics really wants to seewhat's the limits of what's possible.And so, the cost comessecond, or it's a component,but first figure outwhat are the limitations.I think, with Elon, he'sreally driving the cost down.Is there some inspiration,some lessons you see thereof the challenge of driving the cost down,especially with Atlaswith a humanoid robot?- Well, I think the thing thathe's certainly been learningby building car factories iswhat that looks like in scaling.By scaling, you can getefficiencies that drive costs down-- Sure.- very well.And the smart thingthat, you know, they havein their favor is, you know,they know how to manufacture.They know how to build electric motors.They know how to build, you know,computers and vision systems,so there's a lot of overlapbetween modern automotivecompanies and robots.But hey, we have a modern robotic, I mean,automotive company behind us as well.(Lex laughs)- So, bring it on.- Who's doing pretty well, right?The electric vehicles fromHyundai are doing pretty well.- I love it.So, we've talked about someof the low-level control,some of the incrediblestuff that's going onand basic perception, buthow much do you see currentlyand in the future ofBoston Dynamics's sortof higher-level machinelearning applications?Do you see customers addingon those capabilities,or do you see BostonDynamics doing that in house?- Some kinds of thingswe really believe are probably gonna bemore broadly available,maybe even commoditized,you know, using a machine learning,like a vision algorithmso a robot can recognizesomething in the environment.That ought to be somethingyou can just download.Like, I'm going to a newenvironment, and I have a new kindof door handle or piece ofequipment I wanna inspect.You ought to be ableto just download that.And I think peoplebesides Boston Dynamics will provide that.And we've actually builtan API that lets people addthese vision algorithms to Spot,and we're currentlyworking with some partnerswho are providing that.Levatas is a example of a small providerwho's giving us softwarefor reading gauges,and actually, another partnerin Europe, Reply, is doing the same thing.So, we see ultimately an ecosystemof providers doing stuff like that.I think ultimately you might even be ableto do the same thing with behaviors.So, this technology willalso be brought to bearon controlling the robot,the motions of the robot.And you know, we're usinglearning, reinforcement learningto develop algorithms for bothlocomotion and manipulation.And ultimately, this is gonna meanyou can add new behaviors toa robot, you know, quickly.And that could potentially be doneoutside of Boston Dynamics.Right now, that's all internal to us.I think you need to understandat a deep level, you know,the robot control to do that.But eventually, that could be outside.But it's certainly a placewhere these approachesare gonna be broughtto bear in robotics.- So, reinforcement learningis part of the process.So, you do use reinforcement learning.- Yes.(Lex sighs)So, there's increasing levelsof learning with these robots?- Yes.- And that's for locomotion,for manipulation,for perception?- Yes.- Well, what do you think in generalabout all the exciting advancementsof transformer neural networks,most beautifully illustratedthrough the large languagemodels like GPT 4?- Like everybody else,we're all, you know,I'm surprised at how far they've come.I'm a little bit nervous about the,there's anxiety around them, obviously,for, I think, good reasons, right?Disinformation is a curse,an unintended consequenceof social media that could beexacerbated with these tools.So, if you use them todeploy disinformation,it could be a real risk.But I also think that the risksassociated with these kindsof models don't have a whole lot to dowith the way we're gonnause them in our robots.If I'm using a robot, I'mbuilding a robot to do, you know,a manual task of some sort.I can judge very easily, is itdoing the task I asked it to?Is it doing it correctly?There's sort of a built-inmechanism for judging.Is it doing the right thing?Did it successfully do the task?- Yeah, physical realityis a good verifier.- It's a good verifier.That's exactly it, andwhereas if you're askingfor, yeah, I don't know,you're trying to ask atheoretical question in ChatGPT,it could be true, or it may not be true.And it's hard to have that verifier,what is that truth (laughs)that you're comparing against,whereas, in physicalreality, you know the truth.And this is an important difference.And so, I think there is reasonto be a little bit concernedabout, you know, how these tools,large language models could be used.But I'm not very worried abouthow they're gonna be used,well, how learning algorithmsin general are goingto be used on robotics.It's really a differentapplication that has different waysof verifying what's going on.- Well, the nice thingabout language models isthat I ultimately see, I'm really excitedabout the possibility ofhaving conversations with Spot.- Yeah.- There's no, I would say,negative consequences to thatbut just increasing thebandwidth and the varietyof ways you can communicatewith this particular robot.- Yeah.- So, you could communicate visually.You can communicate throughsome interface and to be ableto communicate verballyagain with a beer and so on.I think that's really exciting to makethat much, much easier.- We have this partnerLevatas that's addingthe vision algorithmsfor gauge reading for us.Just this week I saw a demowhere they hooked up, you know,a language tool to Spot,and they're talking to Spotto give commands.- Nice, I love it.- Yeah.- Can you tell me about theBoston Dynamics AI Institute?What is it, and what is its mission?- So, it's a separate organization,the Boston Dynamics ArtificialIntelligence Institute.It's led by Marc Raibert, thefounder of Boston Dynamics,and the former CEO, andmy old advisor at MIT.Marc has always loved the research,the pure research without the confinementor demands of commercialization.And he wanted to continueto, you know, pursue thatunadulterated researchand so suggested to Hyundaithat he set up this institute,and they agree that it'sworth additional investmentto kinda continue pushing this forefront.And we expect to be working togetherwhere you know Boston Dynamics is, again,both commercialize and do research,but the sort of time horizonof the research we'regonna do is, you know,in the next, let's sayfive years, you know.What can we do in the next five years?Let's work on those problems.And I think the goalof the AI Institute isto work even further out.Certainly, you know, the analogyof legged locomotion again,when we started that, thatwas a multi-decade problem.And so, I think Marcwants to have the freedomto pursue really hardover-the-horizon problems.That'll be the goal of the institute.- So, we mentioned some of thedangers, some of the concernsabout large language models.That said, you know, there'sbeen a long-running fearof these embodied robots.Why do you think people are afraid(Robert laughs)of legged robots?- Yeah, I wanted to show you this.So, this is in the Wall Street Journal,and this is all about ChatGPT, right?But look at the picture.- Yeah.- It's a humanoid robot.- That's saying, \"I will replace you.\"- That looks scary, and it says,\"I'm gonna replace you.\"- Yeah.- And so, the humanoid robot isthe embodiment of this ChatGPT toolthat there's reason tobe a little bit nervousabout how it gets deployed.- Yeah.- So, I'm nervous about that connection.It's unfortunate thatthey chose to use a robotas that embodiment.As you and I just said, there'sbig differences in this.But people are afraid becausewe've been taught to be afraidfor over 100 years.So, you know, the word robot was developedby a playwright named Karel Capek in 1921,a Czech playwright,\"Rossum's Universal Robots.\"And in that first depiction of a robot,the robots took over (laughs)at the end of the story.And you know, people love to be afraid.And so, we've beenentertained by these storiesfor 100 years, and I think that's as muchwhy people are afraid as anything else,as we've been sort of taught that this isthe logical progression through fiction.I think it's fiction.- I think what people moreand more will realize,just like you said, that the threat,like say you have asuper-intelligent AI embodiedin a robot.That's much less threateningbecause it's visible.It's verifiable.It's right there in physical reality.And we humans know how todeal with physical reality.I think it's much scarier whenyou have arbitrary scalingof intelligent AI systemsin the digital spacethat they could pretend to be human.So, robot Spot is not gonna pretend.It could pretend it's human all it wants.(Lex laughs)You could put ChatGPT on top of it,but you're gonna know it's not humanbecause you have a contactwith physical reality.- And you're gonna knowwhether or not it's doingwhat you asked it to do.- Yeah, like, it's not gonna, (laughs)I mean, I'm sure it can start,just like a dog lies to you.It's like, \"I wasn't partof tearing up that couch.\"So, Spot can try(Robert laughs)to lie that like, you know,\"It wasn't me that's spilled that thing,\"but you're going to kindof figure it out eventuallyif it happens multiple times, you know.But I think that-- Humanity has figured outhow to make machines safe.- Yeah.- And there's, you know,the regulatory environmentsand certification protocolsthat we've developed in orderto figure out how to make machines safe.We don't know and don't havethat experience with softwarethat can be propagatedworldwide in an instant.And so, I think we neededto develop those protocolsand those tools, and sothat's work to be done.But I don't think the fear of thatand that work shouldnecessarily impede our abilityto now get robots out because again,I think we can judge whena robot's being safe.- So, and again, just like in that image,there's a fear thatrobots will take our jobs.I took a ride.I was in San Francisco.I took a ride in a Waymo vehicle.It's an autonomous vehicle,and I've done it several times.They're doing incredible work over there,but (laughs) people flicked it off.- Oh, really?- Flicked off the car.So, (laughs) I mean, that's a long storyof what the psychology of that is.It could be maybe big tech,or I don't know exactlywhat they're flicking off.- Yeah.- But there is an element of, like,\"These robots are taking our jobs,\"or irreversibly transforming society suchthat it will have economic impact,and the little guy would lose a lot,would lose their well-being.Is there something tobe said about the fearthat robots will take our jobs?- You know,at every significanttechnological transformation,there's been fear of, youknow, an automation anxiety-- Yes.- that it's gonna havea broader impact than we expected.And there will be, youknow, jobs will change.Sometime in the future, we'regonna look back at peoplewho manually unloadedthese boxes from trailers,and we're gonna say, \"Why didwe ever do that manually?\"But there's a lotta peoplewho are doing that job todaythat could be impacted.But I think the realityis, as I said before,we're gonna build the technologyso that those very samepeople can operate it.And so, I think there's a pathwayto upskilling and operating.Just like, look, we usedto farm with hand tools,and now we farm with machines,and nobody has reallyregretted that transformation.And I think the same can besaid for a lot of manual laborthat we're doing today.And on top of that, you know, look,we're entering a new world wheredemographics are gonna havestrong impact on economic growth,and you know, the advanced,the first world is losingpopulation quickly.In Europe, they're worriedabout hiring enough peoplejust to keep the logisticssupply chain going.And you know, part of thisis the response to COVID,and everybody's sort of thinking backwhat they really wanna do with their life,but these jobs are gettingharder and harder to fill.And I'm hearing that over and over again.So, I think, frankly, thisis the right technologyat the right time where we'regonna need some of this workto be done, and we're gonna want toolsto enhance that productivity.- And the scary impact, I think, again,GPT comes to the rescue in termsof being much more terrifying.(Robert laughs)(Lex laughs)The scary impact of, basically,so I'm, I guess, a softwareperson, so I program a lot.And the fact that people likeme could be easily replacedby GPT, that's going to have a-- Well, and lot, you know,anyone who deals with textsand writing a draft proposalmight be easily donewith ChatGPT now.- Yeah.- where-- Consultants.- it wasn't before.- Journalists.- Yeah.- Everybody is sweating.- But on the other hand, youalso want it to be right.And they don't know howto make it right yet.But it might make a good startingpoint for you to iterate.- Boy, do I have to talk toyou about modern journalism.(Robert laughs)That's another conversation altogether,but yes, more right than the average,the mean journalist, yes.You spearheaded the weaponization letterBoston Dynamics has.Can you describe what that letter statesand the general topic ofthe use of robots in war?- We authored a letter and then gotseveral leading roboticscompanies around the world,including, you know, Unitree in China,and Agility here in the United States,and ANYmal in Europe,and, you know, some othersto cosign a letter that saidwe won't put weapons on our robots.And part of the motivationthere is, you know,as these robots start tobecome commercially available,you can see videos online ofpeople who've gotten a robot,and strapped a gun on it, andshown that they can, you know,operate the gun remotely whiledriving the robot around.And so, having a robot thathas this level of mobilityand that can easily be configured in a waythat could harm somebodyfrom a remote operator isjustifiably a scary thing.And so, we felt like it was importantto draw a bright line there and say,\"We're not going to allow this,\"for, you know, reasons that wethink ultimately it's betterfor the whole industryif it grows in a waywhere robots are ultimatelygoing to help us alland make our lives morefulfilled and productive.But by goodness, you're gonnahave to trust the technology,to let it in.And if you think therobot's gonna harm you,that's gonna impede thegrowth of that industry.So, we thought it wasimportant to draw a bright lineand then publicize that.And our plan is to, youknow, begin to engagewith lawmakers and regulators.Let's figure out whatthe rules are going to bearound the use of thistechnology and use our positionas leaders in this industry and technologyto help force that issue.In fact, I have a policy,you know, directorat my company whose job itis to engage with the public,to engage with interestedparties, including regulators,to sort of begin these discussions.- Yeah, it's a really important topic,and it's an importanttopic for people that worryabout the impact of robots on our societywith autonomous weapon systems.So, I'm glad you're sortof leading the way in this.You are the CEO of Boston Dynamics.What's it take to be aCEO of a robotics company?So, you started as ahumble engineer, (laughs)a PhD.Just looking at your journey,what does it take to gofrom building the thingto leading a company?What are some of thebig challenges for you?- Courage I would put front andcenter for multiple reasons.I talked earlier about thecourage to tackle hard problems.So, I think there's couragerequired not just of mebut of all of the peoplewho work at Boston Dynamics.I also think we have a lotof really smart people.We have people who areway smarter than I am.And it takes a kinda courageto be willing to lead themand to trust that you havesomething to offer to somebodywho probably is maybe abetter engineer than I am.Adaptability, you know, it'sbeen a great career for me.I never would've guessed I'd stayedin one place for 30 years, andthe job has always changed.I didn't really aspire to beCEO from the very beginning,but it was the naturalprogression of things.There always needed to be some levelof management that was needed.And so, you know, when I saw somethingthat needed to be donethat wasn't being done,I just stepped in to go do it.And oftentimes because we were fullof such strong engineers,oftentimes that wasin the management direction,or it was in the businessdevelopment directionor organizational, hiring.Geez, I was the main personhiring at Boston Dynamicsfor probably 20 years, so Iwas the head of HR, basically.You know, just willingnessto sort of tackle any pieceof the business that needsit and be willing to shift.- Is there something youcould say to what it takesto hire a great team?What's a good interview process?How do you know the guy or galare gonna make a great memberof a engineering team that's doing someof the hardest work in the world?- You know, we developedan interview processthat I was quite fond of.It's a little bit of ahard interview processbecause the bestinterviews, you ask somebodyabout what they're interestedin and what they're good at,and if they can describe to you somethingthat they worked on, and yousaw they really did the work,they solved the problems, andyou saw their passion for it,but what makes that hard is you haveto ask a probing question about it.You have to be smart enoughabout what they're telling youthey're expert at to ask a good question.And so, it takes a prettytalented team to do that.But if you can do that,that's how you tap into,\"Ah, this person cares about their work.They really did the work.They're excited about it.\"That's the kind of personI want at my company.You know, at Google, they taught usabout their interview process,and it was a little bit different.You know, we evolved theprocess at Boston Dynamicswhere it didn't matterif you were an engineer,or you were an administrative assistant,or a financial person, or a technician.You gave us a presentation.You came in, and yougave us a presentation.You had to stand up andtalk in front of us.And I just thought that wasgreat to tap into those thingsI just described to you.At Google, they taught us,and I understand why, right.They're hiring tens ofthousands of people.They need a more standardized process.So, they would sort oferr on the other sidewhere they would askyou a standard question.I'm gonna ask you a programming question,and I'm just gonna ask youto, you know, write code in front of me.That's a terrifying, youknow, application process.- Yeah.- It does let you comparecandidates really well,but it doesn't necessarilylet you tap into who they are.- Yeah.- (laughs) Right?'Cause you're asking themto answer your questioninstead of you asking them aboutwhat they're interested in.But frankly, thatprocess is hard to scale.And even at Boston Dynamics,we're not doing thatwith everybody anymore.But we are still doing thatwith, you know, the technical peoplebecause we too now need to sortof increase our rate of hiring.Not everybody's givinga presentation anymore.- But you're stillultimately trying to findthat basic seed of passion-- Yeah, and talent.- for the world.- You know, did they really do it?Did they find somethinginteresting or curious, you know,and do they care about it? (laughs)- I think somebody I admire is Jim Keller,and he likes details.So, one of the ways you could, (laughs)if you get a person to talkabout what they're interestedin, how many details, like, how muchof the whiteboard can you fill out?- Yeah.- What they-- Well, I think you figure outdid they really do the workif they know some of the details.- Yes.- And if they haveto wash over the details,well, then they didn't do it.- They didn't do it.(Robert laughs)'Cause especially with engineering,the work is in the details.- Yeah.- I have to go there briefly (sighs)just to get your kind of thoughtson the long-term future of robotics.There's been discussions on the GPT side,on the large language modelside of whether there'sconsciousness insidethese language models.And I think there'sfear, but I think there'salso excitement or at least the wide worldof opportunity andpossibility in embodied robotshaving something like,let's start with emotion,love towards other human beingsand perhaps the display, realor fake, of consciousness.Is this something you think about in termsof long-term future?Because, as we've talked about,people do anthropomorphize these robots.It's difficult not to project some levelof, I use the word sentience,some level of sovereignty, identity,all the things we think as human.That's what anthropomorphizationis, is we project humannessonto mobile, especially legged robots.Is that something almostfrom a science-fictionperspective you think about?Or, do you try to avoid ever,try to avoid the topic ofconsciousness altogether?- I'm certainly not an expert in it,(Lex laughs)and I don't spend-- Is anybody?- a lot of time thinkingabout this, right?And I do think it's fairlyremote for the machinesthat we're dealing with.You're right that people anthropomorphize.They read into the robots'intelligence and emotionthat isn't there becausethey see physical gesturesthat are similar tothings they might even seein people or animals.I don't know muchabout how these largelanguage models really work.I believe it's a kindof statistical averagingof the most common responses, you know,to a series of words, right?It's sort of a veryelaborate word completion.And I'm dubious that that hasanything to do with consciousness.And I even wonder if that model of sortof simulating consciousnessby stringing words togetherthat are statisticallyassociated with one another,whether or not that kind of knowledge,if you wanna call that knowledge,would be the kind of knowledgethat allowed a sentient beingto grow or evolve.It feels to me like there'ssomething about truthor emotions that's just a verydifferent kind of knowledgethat is absolute.The interesting thing abouttruth is it's absolute,and it doesn't matter howfrequently it's representedin the worldwide web.If you know it to betrue, it it can only be,it may only be there once,but by God, that's true.And I think emotions are alittle bit like that, too.You know something, you know,and I just think that's adifferent kind of knowledgethan the way these largelanguage models derive sortof simulated-- It does seem that-- intelligence.- things that are true very well might bestatistically wellrepresented on the Internetbecause the Internet's made up of humans.So, I tend to suspect thatlarge language models are goingto be able to simulateconsciousness very effectively.And I actually believe that current GPT 4,when fine-tuned correctly,would be able to do just that.And there's going to be a lotof very complicatedethical questions that haveto be dealt with that havenothing to do with roboticsand everything to do with-- There needs to be someprocess of labeling,I think, (laughs) what is truebecause there is alsodisinformation available on the web,and these models are goingto consider that kindof information as well.And again, you can't averagesomething that's trueand something that'suntrue and get somethingthat's moderately true. (laughs)It's either right, or it's wrong.And so, how is that process,and this is obviously somethingthat the purveyors ofthese, Bard and ChatGPT,I'm sure this is what they're working on.- Well, if you interact onsome controversial topicswith these models, they'reactually refreshingly nuanced.Well, you realize there'sno one truth, you know.What caused the war in Ukraine, right?Any geopolitical conflict, youcan ask any kind of question,especially the ones thatare politically tense,divisive, and so on.GPT is very good at presenting,it presents the different hypotheses.It presents calmly(laughing) sort of the amountof evidence for each one.It's really refreshing.It makes you realizethat truth is nuanced,and it does that well.And I think, with consciousness,it would very accurately say,\"Well, it sure as hell feelslike I'm one of you humans,but where's my body?(Robert laughs)I don't understand.\"Like, you're going to be confused.The cool thing about GPT isit seems to be easily confusedin the way we are.Like, you wake up in a newroom, and you ask, \"Where am I?\"It seems to be able todo that extremely well.It'll tell you one thing, like a factabout when a war started,and when you correct it, say,\"Well, that's not consistent,\"it'll be confused.It'll be, \"Yeah, you're right.\"It'll have that sameelement, childlike elementwith humility of trying tofigure out its way in the world.And I think that's a really tricky areato sort of figure out withus humans of what we wantto allow AI systems to say to us.Because then, if there'selements of sentiencethat are on display, you can then startto manipulate human emotion,all that kinda stuff.But I think that's a really seriousand aggressive discussionthat needs to be had(laughing) on the software side.I think, again, embodiment,robotics are actually saving usfrom the arbitrary scalingof software systemsversus creating more problems.But that said, I really believein that connectionbetween human and robot.There's magic there.And I think there's also, I think,a lot of money to be made there.And Boston Dynamics is leading the worldin the most elegant movementdone by robots.(Robert laughs)So, I can't wait-- Well, thank you.- to what maybe otherpeople that built on topof Boston Dynamics robots orBoston Dynamics by itself.So, you had one wild career,one place on one set of problemsbut incredibly successful.Can you give advice to youngfolks today in high school,maybe in college lookingout into this futurewhere so much roboticsand AI seems to be definingthe trajectory of human civilization.Can you give 'em adviceon how to have a careerthey can be proud of or how to havea life they can be proud of?- Well, I would say, youknow, follow your heartand your interest.Again, this was an organizingprinciple, I think,behind the Leg Lab at MIT that turnedinto a value at Boston Dynamics,which was follow your curiosity.Love what you're doing.You'll have a lot more fun,and you'll be a lot betterat it as a result.I think it's hard to plan, you know.Don't get too hung up onplanning too far ahead.Find things that you like doingand then see where it takes you.You can always change direction.You will find things that, you know,\"Ah, that wasn't a good move.I'm gonna pack up andgo do something else.\"So, when people aretrying to plan a career,I always feel like, \"Ah,there's a few happy mistakesthat happen along the wayand just live with that it.\"You know, but make choices then.So, avail yourselves to theseinteresting opportunities,like when I happened to runinto Marc down in the lab,the basement of the AI lab, but be willingto make a decision and then pivotif you see somethingexciting to go at, you know,'cause, if you're out and about enough,you'll find things likethat that get you excited.- So, there was a feelingwhen you first met Marcand saw the robots thatthere's something interesting.- \"Oh, boy, I gotta go do this.\"There was no doubt.(Lex laughs)(Robert laughs)- What do you think in 100 years, whoo,what do you think BostonDynamics is doing?Even bigger, what do you think is the roleof robots in society?Do you think we'll be seeingbillions of robots everywhere?Do you think about that long-term vision?- Well, I do think that,I think that robots will be ubiquitous,and they will be out amongst us,and they'll be certainly doing, you know,some of the hard labor that we do today.I don't think people don't wanna work.People wanna work.People need to work to,I think, feel productive.We don't wanna offload allof the work to the robots'cause I'm not sure if people would knowwhat to do with themselves.(Lex laughs)And I think just self-satisfactionand feeling productive issuch an ingrained part of being humanthat we need to keep doing this work.So, we're definitely gonna have to workin a complimentary fashion,and I hope that the robotsand the computers don't end up being ableto do all the creative work.Right?- Yeah.- 'Cause that's the part that's, you know,that's the rewarding.The creative part of solvinga problem is the thingthat gives you that serotoninrush that you never forget,you know, (laughs) or that adrenaline rushthat you never forget, andso, you know, people needto be able to do that creative workand just feel productive,and sometimes you can feel productiveover fairly simple workthat's just well done,you know, and that youcan see the result of.So, yeah, you know, I don't know,there was a cartoon, was it \"Wall-E,\"where they had this big ship,and all the people were just overweight,lying on their beach chairs kinda slidingaround on the deck of the moviebecause they didn't do anything anymore.- Yeah.- Well,we definitely don't wannabe there, (laughs) you know.We need to work in somecomplimentary fashionwhere we keep all of ourfaculties and our physical health,and we're doing some labor, right,but in a complimentary fashion somehow.- And I think a lotta that hasto do with the interaction,the collaboration withrobots and with AI systems.I'm hoping there's a lot ofinteresting possibilities there.- I think that couldbe really cool, right?If you can work in an interactionand really be helpful,robots, you know, you canask a robot to do a jobyou wouldn't ask a person to do,and that would be a real asset.You wouldn't feel guiltyabout it, you know. (laughs)You'd say, \"Just do it.\"- Yeah.- It's a machine.And I don't have to havequalms about that, you know.- The ones that are machines,I also hope to see a future,and it is hope.I do have optimism about the futurewhere some of the robots are pets,have an emotional connection to us humansand because one of the problemsthat humans have to solve isthis kind of general loneliness.The more love you have in your life,the more friends you have in your life,I think that makes a moreenriching life, helps you grow.And I don't fundamentally see why someof those friends can't be robots.- There's an interestinglong-running study,maybe it's in Harvard, justnice report article writtenabout it recently.They've been studying this groupof a few thousand peoplenow for 70 or 80 years.And the conclusion is thatcompanionship and friendship arethe things that make fora better and happier life.And so, I agree with you,and I think that couldhappen with a machinethat is probably, you know,simulating intelligence.I'm not convinced there willever be true intelligencein these machines, sentience.But they could simulate it,and they could collect your history.You know, I guess it remains to be seenwhether they can establish that real deep,you know, when you sit with a friend,and they remember somethingabout you and bring that up,and you feel that connection,it remains to be seenif a machine's gonna beable to do that for you.- Well, I have to say, inklingsof that already started happening for me.Some of my best friends are robots.(Robert laughs)And I have you to thankfor leading the wayin the accessibility, and theease of use of such robots,and the elegance of their movement.Robert, you're an incredible person.Boston Dynamics is an incredible company.I've just been a fan for many, many yearsfor everything you stand for,for everything you do in the world.If you're interested ingreat engineering, robotics,go join them.Build cool stuff.I'll forever celebratethe work you're doing,and it's just a big honorthat you would sit with me today and talk.It means a lot, so thank you so much.Keep doing great work.- Thank you, Lex.I'm honored to be here,and I appreciate it.It was fun.- Thanks for listeningto this conversationwith Robert Playter.To support this podcast, please checkout our sponsors in the description.And now, let me leave you some wordsfrom Alan Turing in 1950,defining what is nowtermed the Turing test.\"A computer would deserveto be called intelligentif it could deceive a humaninto believing that it was human.\"Thank you for listening andhope to see you next time.\n"