Skip to content
Vijay Kumar: Flying Robots | Lex Fridman Podcast #37
This video isn’t embeddableWatch on YouTube →
Lex Fridman PodcastLex Fridman Podcast

Vijay Kumar: Flying Robots | Lex Fridman Podcast #37

Lex Fridman and Vijay Kumar on vijay Kumar reveals future of agile flying robots and swarms.

Lex FridmanhostVijay Kumarguest
Sep 8, 201956mWatch on YouTube ↗

EVERY SPOKEN WORD

  1. 0:00 – 3:37

    Building a 7,000-pound hexapod: early lessons in coordination

    1. LF

      The following is a conversation with Vijay Kumar. He's one of the top roboticists in the world, a professor at the University of Pennsylvania, a dean of Penn Engineering, former director of GRASP Lab, or the General Robotics Automation Sensing and Perception Laboratory at Penn that was established back in 1979. That's 40 years ago. Vijay is perhaps best known for his work in multi-robot systems, robot swarms, and micro aerial vehicles, robots that elegantly cooperate in flight under all the uncertainty and challenges that the real-world conditions present. This is the Artificial Intelligence Podcast. If you enjoy it, subscribe on YouTube, give it five stars on iTunes, support it on Patreon, or simply connect with me on Twitter at Lex Fridman, spelled F-R-I-D-M-A-N. And now, here's my conversation with Vijay Kumar. What is the first robot you've ever built or were a part of building?

    2. VK

      Way back when I was in graduate school, I was part of a fairly big project that involved building a very large hexapod. This weighed close to 7,000 pounds, and it was powered by hydraulic actuation, or was actuated by hydraulics with 18 motors, hydraulic motors, each controlled by an Intel 8085, uh, processor and an Inte- 8086 co-processor. And so imagine this huge, uh, monster that had 18 joints, each controlled by an independent computer, and there was a 19th computer that actually did the coordination between these 18 joints. So I was part of this project and my thesis work was how do you coordinate the 18 legs, and in particular, the-the pressures in the hydraulic cylinders to get efficient locomotion.

    3. LF

      It sounds like a giant mess. So how difficult is it to make all the motors communicate? Presumably, you have to send signals hundreds of times a second, or at least-

    4. VK

      Yeah, so this was not my work, but the-the folks who worked on this wrote what I believe to be the first multiprocessor operating system. This was-

    5. LF

      Mm-hmm.

    6. VK

      ... in the '80s. And you have to make sure that, uh, obviously messages got across from one joint to another. You have to remember, the-the clock speeds on those computers were about half a megahertz.

    7. LF

      (laughs) Right.

    8. VK

      So.

    9. LF

      The '80s. So not to romanticize the notion, but how did it make you feel to make-to see that robot move?

    10. VK

      It was amazing. In hindsight, it looks like, well, we built this thing which really should have been much smaller. And of course, today's robots are much smaller. You look at, you know, Boston Dynamics or Ghost Robotics, a spinoff from-from Penn. But back then, you were stuck with the substrate you had, the compute you had, so things were unnecessarily big. But at the same time, uh, and this is just human psychology, somehow bigger means grander. You know, people never have the same appreciation for nanotechnology or nano devices as they do for the space shuttle or the Boeing 747.

    11. LF

      Yeah, you've actually done quite a good job at illustrating that small is beautiful.

    12. VK

      Yeah.

    13. LF

      In terms of robotics. (laughs)

    14. VK

      Yeah.

  2. 3:37 – 5:32

    Why small aerial robots are “beautiful”: formation flight and 3D shapes

    1. LF

      So what is, uh, on that topic is the most beautiful or elegant robot in motion that you've ever seen? Not to pick favorites or whatever, but something that just inspires you, that you remember.

    2. VK

      Well, I think thing that I'm-I'm most proud of that my students have done is really think about, uh, small UAVs that can maneuver in constrained spaces, and, uh, p- in particular, their ability to coordinate with each other and form three-dimensional patterns. So once you can do that, uh, you can, uh, essentially create 3D objects in the sky, and you can deform these objects on the fly. So in some sense, your toolbox of what you can create is suddenly got enhanced.

    3. LF

      Mm-hmm.

    4. VK

      And before that, we did the two-dimensional version of this, so we had, uh, ground robots forming patterns and-and so on. So that-that was not as impressive, that was not as beautiful. But if you do it in 3D, suspended in midair, and you've got to go back to 2011 when we did this, um, now it's actually pretty standard to do these things eight years later, um, but back then, it was a big accomplishment.

    5. LF

      So the distributed cooperation is where-is where beauty emerges, in your eyes?

    6. VK

      Well, I think beauty to an engineer is very different from-from beauty to, you know, someone who's looking at robots from the outside, if you will.

    7. LF

      Yeah.

    8. VK

      But what I meant there, so before we said that grand is associated with size, and, um, another way of thinking about this is just the physical shape, and the idea that you can create physical shapes in midair and have them deform, uh, that's beautiful.

    9. LF

      But the individual components, the agility is beautiful too, right?

    10. VK

      That is true too.

    11. LF

      So the-the aggressive move.

    12. VK

      So then how-how quickly can you actually manipulate these three-dimensional shapes and the individual components? Yes, you're right.

  3. 5:32 – 6:23

    Drones vs aerial robots: language, autonomy, and agility

    1. LF

      Uh, by the way, you said UAV, unmanned aerial vehicle. Well, what's a good term for drones, UAVs, quadcopters? Is there a term that's stan- being standardized?

    2. VK

      I don't know if there is. Everybody wants to use the word drones. And I've often said this, drones to me is a pejorative word. It-it signifies something that's, uh, that's dumb-

    3. LF

      Hmm.

    4. VK

      ... that's pre-programmed, that does one little thing. And robots are anything but drones. So I actually don't like that word, but that's what everybody uses.You could call it unpiloted.

    5. LF

      Unpiloted.

    6. VK

      Uh, but even unpiloted, could be radio controlled, could be remotely controlled in many different ways. And I think the right word is thinking about it as an aerial robot.

    7. LF

      You also say agile, autonomous, aerial robot, right?

    8. VK

      Yeah. So agility is an attribute, but they don't have to be.

  4. 6:23 – 8:33

    Ants and emergence: robustness, consensus, and swarm intelligence

    1. LF

      So what biological system? 'Cause you've also drawn a lot of inspiration with those. I've seen bees and ants that you've talked about. Uh, what living creatures have you found to be most inspiring as an engineer, instructive in your work in robotics?

    2. VK

      To me, so ants are, are really, um, uh, quite incredible creatures, right? So you, I mean, the individuals, uh, arguably are s- very simple in how they're, they're built, uh, and yet they're incredibly resilient as a population, and as individuals, they're incredibly robust. So, you know, if you take an ant, uh, with six legs, you remove one leg, it still works just fine. And, uh, it moves along and I don't know that he even realizes it's lost a leg. So that's the robustness of the individual ant level.

    3. LF

      Mm-hmm.

    4. VK

      Uh, but then you, you look about this instinct for self-preservation of the colonies and they adapt in so many amazing ways, you know, transcending, transcending gaps, um, and, and, by, by just chaining themselves together when you have a flood-

    5. LF

      Mm-hmm.

    6. VK

      ... uh, being able to recruit other teammates to carry big morsels of food, and then going out in different directions looking for food, and then being able to demonstrate consensus, uh, even though they don't communicate directly with each other the way we communicate with each other.

    7. LF

      Mm-hmm.

    8. VK

      Um, in some sense, they also know how to do democracy-

    9. LF

      (laughs)

    10. VK

      ... probably better than what we do.

    11. LF

      Yeah. Somehow it's the, even democracy is emergent. It seems like all of the phenomena that we see is all emergent. It seems like there's no centralized communicator.

    12. VK

      There is. So the, I think a lot is made about that word, uh, emergent, and it means lots of things to different people, but you're absolutely right. I think, as an engineer, you think about what element, elemental behaviors or perimeters you could synthesize so that the whole looks incredibly powerful, incredibly synergistic, the whole definitely being greater than the sum of the parts. And ants are living proof of that.

  5. 8:33 – 11:16

    Scaling swarms: abstraction, interfaces, and resiliency under failures

    1. LF

      So, uh, when you see these beautiful swarms, whether it's biological systems of robots, do you sometimes think of them as a single individual living intelligent organism? So it's the same as thinking of our human civilization as one organism? Or do you still, as an engineer, think about the individual components and all the engineering that went into the individual components?

    2. VK

      Well, that's very interesting. So again, philosophically, as engineers, what we wanna do is to go beyond the individual components, the individual units-

    3. LF

      Mm-hmm.

    4. VK

      ... and think about it as a unit, as a cohesive unit without worrying about the individual components. If you start obsessing about the individual building blocks and what they do, uh, you inevitably will find it hard to scale up. Just mathematically, just think about individual things you wanna model, and if you want to have 10 of those, then you essentially are taking Cartesian products of 10 things. That makes it really complicated. Then to do any kind of synthesis or design in that high dimensional space is really hard.

    5. LF

      Mm-hmm.

    6. VK

      So the right way to do this is to think about the individuals in a clever way so that, at the higher level, when you look at lots and lots of them, abstractly, you can think of them in some low dimensional space.

    7. LF

      So what, what does that involve? You, for the individual, you have to try to make the way they see the world as local as possible? And the other thing, do you just have to make 'em robust to collisions, like you said with the ants, if something fails, the, the, the whole swarm doesn't fail?

    8. VK

      Right. I, I think as, as engineers, we do this. I mean, you know, think about we build planes or we build iPhones. Um, and we know that by taking individual components, well-engineered components with well-specified interfaces-

    9. LF

      Yeah.

    10. VK

      ... that behave in a predictable way, you can build complex systems. Um, so that's ingrained, I would, I would claim, in most engineers' thinking, um, and it's true for computer scientists as well. I think what's different here is that you want, uh, the individuals, uh, to be robust in some sense, um, as we do in these other settings. But you also want some degree of resiliency for the population. And so you really want them to be able to reestablish, uh, communication with their neighbors. You want them to rethink their strategy for group behavior. You want them to reorganize, um, and that's where I think a lot of the challenges lie.

  6. 11:16 – 15:04

    Nature vs engineered swarms: missions, global frames, and mapping

    1. LF

      So just, uh, at a high level, what does it take for a bunch of, uh, what should we call them, flying robots, to create a formation? Just, uh, at a, for people who are not familiar with r- with robotics in general, how much information is needed? How do you, how do you even make it happen without a centralized controller?

    2. VK

      So, I mean, there are a couple of different ways of looking at this. If you are a purist, uh, you think of it as a, uh, as a, as a way of recreating what nature does. So nature forms groups for several reasons, but mostly it's because of this instinct that organisms have of preserving their colonies, their population.... which means what? You need shelter. You need food. You need to procreate, and that's basically it. So the kinds of interactions you see are all organic, they're all local. Um, and the only information that they share, and mostly it's indirectly, is to, again, preserve the herd, or the flock-

    3. LF

      Mm-hmm.

    4. VK

      ... uh, or the swarm, uh, in, and either by looking for new sources of food or looking for new shelters, right?

    5. LF

      Right.

    6. VK

      Um, as engineers, when we build swarms, we have a mission And when you think of a mission and it involves mobility, most often it's described in some kind of a global coordinate system.

    7. LF

      Mm-hmm.

    8. VK

      As a human, as an operator, as a commander, or as a collaborator, I have my coordinate system-

    9. LF

      Okay.

    10. VK

      ... and I want the robots to be consistent with that.

    11. LF

      Right.

    12. VK

      So I might think of it slightly differently. I might want the robots to recognize that coordinate system, uh, which means not only do they have to think locally in terms of who their immediate neighbors are, but they have to be cognizant of, of what the global environment looks like. So if I go, if I say, "Surround this building and protect this from intruders," well, they're immediately in a building-centered coordinate system, and I have to tell them where the building is.

    13. LF

      And they're globally collaborating on a map of that building. They're, they're maintaining some kinda global, not just in the frame of the building, but there's information that's ultimately being built up explicitly, as opposed to kind of, uh, uh, implicitly like nature might.

    14. VK

      Correct, correct.

    15. LF

      Yeah.

    16. VK

      So in some sense, nature is very, very sophisticated, but the tasks that nature solves or needs to solve are very different from the kind of engineered tasks, artificial tasks that we are, uh, forced to address. Um, and again, there's nothing preventing us from solving these other problems, but ultimately, it's about impact. You want these swarms to do something useful, um, and so you're kinda driven into this, uh, very unnatural, if you will, unnatural meaning not like how nature does-

    17. LF

      Mm-hmm.

    18. VK

      ... setting.

    19. LF

      And it's a little, probably a little bit more expensive to do it the way nature does, because, uh, nature is, uh, less sensitive to the loss of the individual. And, uh, cost-wise, in robotics, I think you're more sensitive to losing individuals.

    20. VK

      I, I, I think that's true, although, uh, if you look at the price-to-performance ratio of robotic components, it's, it's coming down dramatically.

    21. LF

      Oh, interesting.

    22. VK

      Right? It continues to come down, so I think we're asymptotically approaching-

    23. LF

      (laughs) Nature.

    24. VK

      ... the point where we would get, yeah, the cost of individuals would really become insignificant.

  7. 15:04 – 20:27

    Autonomous flying vehicles and “true autonomy” without infrastructure

    1. LF

      Yeah. So let's step back at a high-level view, the impossible question of what kind of, and as an overview, what kind of autonomous flying vehicles are there, in general?

    2. VK

      I think the ones that, uh, receive a lot of notoriety are obviously the military vehicles. Military vehicles are controlled by a base station, but have, uh, uh, a lot of human supervision, but they have limited in- autonomy, which is the ability to go from point A to point B, and even the more sophisticated now, f- sophisticated vehicles can do autonomous, uh, takeoff and landing.

    3. LF

      And those usually have wings, and they're heavy, and they-

    4. VK

      Usually, they're wings, but then there's nothing preventing us from doing this for helicopters as well. There are, um, I mean, there are many military organizations that have autonomous helicopters in the same vein. And by the way, you, you look at autopilots in airplanes, and it's, it's actually very similar. In fact, I can... One interesting question we can ask is, uh, if you look at all the air safety violations or all the crashes that occurred-

    5. LF

      Yeah.

    6. VK

      ... would they have happened if the plane were truly autonomous? And I think you'll find that, in many of the cases, you know, because of pilot error, we make silly decisions. And so, in some sense, even in air traffic, commercial air traffic, there's a lot of applications, uh, although we only see, um, autonomy being enabled at very high altitudes, uh, when, when the, the, the pilot, the, the, the, the plane is on autopilot.

    7. LF

      There's still a role for the human, and tho- that kind of, uh, autonomy is, uh, you're kind of implying, I don't know what the right word is, but it's a little dumb- dumber than it could be.

    8. VK

      Uh, right.

    9. LF

      It's simpler and naive.

    10. VK

      So, so in the lab, of course, we can, we can, we can afford to be a lot more aggressive.

    11. LF

      Right.

    12. VK

      And the question we try to ask is, um, can we make robots that would be able to make decisions without any kind of external infrastructure?

    13. LF

      Right.

    14. VK

      So what does that mean? So the most common piece of infrastructure that airplanes use today is GPS. Uh, GPS is also the most brittle form of information. Um, if you have driven in a city, tried to use GPS navigation, you know, in tall buildings, you immediately lose GPS. Um, and so that's not a very sophisticated way of building autonomy. I think the second piece of infrastructure they rely on is communications. Again, it's very easy to jam communications. In fact, if you use wifi, you know that wifi signals drop out, cell signals drop out, so to rely on something like that is not, is not good. The third form of infrastructure we, we use, and I hate to call it infrastructure, but, but it is that in the sense of robots, is people.

    15. LF

      (laughs)

    16. VK

      So you could rely on somebody to pilot you. (laughs)

    17. LF

      Right.

    18. VK

      Uh, and so the question you wanna ask is if there are no pilots, if there's no communications with any base station-

    19. LF

      Yeah.

    20. VK

      ... if there's no knowledge of position-... and if there's no a priori map, a priori knowledge-

    21. LF

      Yeah.

    22. VK

      ... of what the environment looks like, a priori model of what might happen in the future, can robots navigate? So that is true autonomy.

    23. LF

      Right. So that's, that's true autonomies. And we're talking about, you mentioned, like military applications of drones. Okay, so what else is there? You talk about agile, autonomous flying robots, aerial robots. So that's a different kind of... It's not winged, it's not big. At least, it's small.

    24. VK

      So I use the word agility mostly, or at least we're motivated to do agile robots, mostly because robots can operate and should be operating in constrained environments.

    25. LF

      Mm-hmm.

    26. VK

      And if you want to operate the way a Global Hawk o- operates, I mean, the kinds of conditions in which you operate are very, very restrictive. If you go, want to go inside a building, for example, for search and rescue, or to locate an active shooter, or you want to navigate under the canopy in an orchard to look at health of plants or to, to look for, count, to count fruits-

    27. LF

      Mm-hmm.

    28. VK

      ... to measure the tree, the tree trunks, uh, these are things we do, by the way. Uh-

    29. LF

      Yeah. Yeah, some cool agriculture stuff you've shown in, in the past. It's really awesome.

    30. VK

      Right. So, so in those kinds of settings, you j- you do need that agility. Agility does not necessarily mean you, you break records for the 100 meters dash. What it really means is you see the unexpected and you're able to maneuver, uh, in a safe way and in a way that, uh, that gets you the most information about the thing you're trying to do.

  8. 20:27 – 20:52

    How quadcopters fly: sensing, control loops, and underactuation

    1. LF

      So, uh, on that point, simply, what does it take to make a thing with four motors fly, a quadcopter, one of these little flying robots? Wh- y- you know, how hard is it to make it fly? How do you coordinate the four motors? What's, how do you convert those, those motors into actual movement?

  9. 20:52 – 27:08

    Why 2007–2009 mattered: IMUs, iPhone-era compute, and commoditization

    1. VK

      So, this is an interesting question. We've been trying to do this since 2000. It is a commentary on the sensors that were available back then, the computers that were available back then. And a number of things happened between 2000 and 2007. One is the advances in computing, which is, and so we all know about Moore's law, but I think, uh, 2007 was a tipping point, the year of the iPhone, the year of the cloud. Uh, lots of things happened in 2007. Um, but going back even further, inertial measurement units as a sensor really matured. Again, lots of reasons for that. Um, certainly there's a lot of federal funding, particularly DARPA in the US, but they didn't anticipate this boom, uh, in IMUs.

    2. LF

      Mm-hmm.

    3. VK

      Um, but if you look, uh, subsequently, what happened is that every air- every car manufacturer had to put an airbag in, which meant you had to have an accelerometer on board. And so that drove down the price to performance ratio of these sensors.

    4. LF

      Oh, I never... So, I should know this.

    5. VK

      Uh-

    6. LF

      That's very interesting.

    7. VK

      Yeah.

    8. LF

      That's very interesting, the connection there.

    9. VK

      S- and that's why research is very s- uh, it's very hard to predict the outcomes. And again, the federal government spent a ton of money on things that they thought were useful for resonators-

    10. LF

      Mm-hmm.

    11. VK

      ... but it ended up enabling the small UAVs-

    12. LF

      Yeah.

    13. VK

      ... uh, which is great 'cause I could have never raised that much money and told, you know, s- sold this project, "Hey, we wanna build these small UAVs. Can you, can you actually fund the development of low-cost IMUs?"

    14. LF

      So, why do you need an IMU on a UAV?

    15. VK

      So, so, so I was, I- I'll come back to that. But, but, so in 2007, 2008, we were able to build these, and then the question you're asking was a good one, how do you coordinate the motors-

    16. LF

      Right.

    17. VK

      ... to, to, to develop this? Uh, but over the last 10 years, everything is commoditized. A high school kid today can pick up a Raspberry Pi, uh, kit, uh, and build this. All the low-levels functionality is all automated. Um, but basically, at some level, uh, you have to drive the motors at the right RPMs, the right, uh, velocity in order to generate the right amount of thrust, in order to position it and orient it in a way that you need to in order to fly. Um, the feedback that you get is from onboard sensors, and the IMU is an important part of it. The IMU tells you what the acceleration is-

    18. LF

      Mm-hmm.

    19. VK

      ... uh, as well as what the angular velocity is-

    20. LF

      Mm-hmm.

    21. VK

      ... uh, and those are important pieces of information. In addition to that, you need some kind of local position or velocity information. For example, when we walk, we implicitly have this information because we kinda know how, how, what our stride length is.

    22. LF

      Mm-hmm.

    23. VK

      Uh, we also are looking at images fly past our retina, if you will-

    24. LF

      Mm-hmm.

    25. VK

      ... and so we can estimate velocity. We also have accelerometers in our head, and we're able to integrate all these pieces of information to determine where we are as we walk. And so robots have to do something very similar. You need an IMU, you need some kind of a camera or other sensor that's measuring velocity-

    26. LF

      Mm-hmm.

    27. VK

      ... um, and then you need some kind of a global reference frame if you really want to think about, um, doing something in a world coordinate system. And so how do you-... estimate your position with respect to that global reference frame, that's important as well.

    28. LF

      So coordinating the RPMs of the four motors is what allows you to, first of all, fly and hover, and then you can change the orientation and the velocity of the... and so on.

    29. VK

      Exactly, exactly, so-

    30. LF

      So it's a bunch of, uh, degrees of freedom that you're playing with.

  10. 27:08 – 29:22

    From point A to point B: trajectory planning for safety and optimality

    1. LF

      Yeah, and you have to send the signal maybe 100 times a second, so the compute there is... everything has to come down in price. And, uh, what are the steps of getting from point A to point B? So y- we just talked about like local control, but if all the kind of cool, uh, dancing in the air that I've seen you show, how do you make it happen? How, y- trajector- make a trajectory... First of all, okay, f-figure out a trajectory, so plan a trajectory, and then how do you make that trajectory happen?

    2. VK

      Yeah, I think planning is a very fundamental problem in robotics. I think, you know, 10 years ago, it was an esoteric thing, but today with self-driving cars, you know, everybody can understand this basic idea that a car sees a whole bunch of things and it has to keep a lane or maybe make a right turn or switch lanes. It has to plan a trajectory.

    3. LF

      Okay.

    4. VK

      It has to be safe, it has to be efficient. So everybody's familiar with that. That's kind of the first step that, that, uh, you have to think about when you, when you, when you, uh, when you say autonomy. Um, and so, uh, for us, it's about finding smooth motions, motions that are safe. Um, so we think about these two things. One is optimality, one is safety. Clearly, you don't, you cannot compromise safety, um, so you're looking for safe, optimal motions. Uh, the other thing you have to think about is can you actually compute a reasonable trajectory f- in a fast amo- in a, in a small amount of time? 'Cause you have a time budget. So the optimal becomes suboptimal.

    5. LF

      Mm-hmm.

    6. VK

      But, um, in our lab, we, we focus on, uh, synthesizing smooth trajectory that satisfy all the constraints, in other words, don't violate any safety constraints, um, and is as efficient as possible. And when I say efficient, it could mean I want to get from point A to point B as quickly as possible or I want to get to it as gracefully as possible, um, or I want to consume as little energy as possible.

    7. LF

      But always staying within the safety constraints.

    8. VK

      But... Yes.

    9. LF

      Okay.

    10. VK

      Always finding a safe, uh, trajectory.

  11. 29:22 – 33:52

    Learning in flight: modeling limits, aerodynamic effects, and hybrid methods

    1. LF

      So there's a lot of excitement and progress in the field of machine learning-

    2. VK

      Yes.

    3. LF

      ... or, and, uh, reinforcement learning and the neural network variant of that with deep reinforcement learning. Do y- do you see a role of machine learning in... So a lot of the success with flying robots did not rely on machine learning except for maybe a little bit of the perception, the computer vision side. On the control side and the planning, do you see there's a role in the future for machine learning?

    4. VK

      So let me, uh, disagree a little bit with you. I think we never perhaps called out in my work, called out learning, but even this very simple idea of being able to fly through a constrained space, the first time you try it, you'll invariably... you might get it wrong-

    5. LF

      Mm-hmm.

    6. VK

      ... if, if the task is challenging. And the reason is, to get it perfectly right, you have to model everything in the environment.

    7. LF

      Right.

    8. VK

      And flying is notoriously hard to model. There are aerodynamic effects that we constantly discover, uh-... even just before I was talking to you, I was talking to a student about how blades flap when they fly.

    9. LF

      Huh. Wow.

    10. VK

      And that ends up changing how a rotorcraft is accelerated in the angular direction.

    11. LF

      So these are like micro flaps or something? It's mo- it's-

    12. VK

      It's not micro flaps. So we assume that each blade is, is rigid but actually it flaps a little bit.

    13. LF

      Oh.

    14. VK

      It bends.

    15. LF

      Interesting. Yeah.

    16. VK

      And so the models rely on the fact, on the, on, on an assumption that they're, they're actually rigid. But that's not true. If you're flying really quickly, these effects become significant. If you're flying close to the ground, you get pushed off by the ground, right? Something which every pilot knows when, when he tries to land or she tries to land.

    17. LF

      Mm-hmm.

    18. VK

      This is, this is called the ground effect. Something very few pilots think about is what happens when you go close to a ceiling or you get sucked into a ceiling. There are very few aircrafts that fly close to any kind of ceiling. Likewise, when you go close to, close to a wall, uh, there are these wall effects. Um, and if you've gone on a train and you pass another train that's traveling the opposite direction, you feel the buffeting. And so these kinds of micro climates-

    19. LF

      Mm-hmm.

    20. VK

      ... affect our UAVs significantly. So if you want-

    21. LF

      And they're impossible to model, essentially (overlapping)

    22. VK

      I wouldn't say they're-

    23. LF

      ... explicitly.

    24. VK

      ... impossible to model, but the level of sophistication you would need in the model and the software would be tremendous. Uh, plus to get everything right would be awfully tedious. So the way we do this is over time, we figure out how to adapt to these conditions. Um, so we've, early on, we used a form of learning that we call iterative learning. Uh, so this idea if you wanna perform a task, um, there are a few things that you need to change-

    25. LF

      Mm-hmm.

    26. VK

      ... uh, and iterate over, few parameters, that over time, you can, you can, you can figure out. So I could call it policy gradient reinforcement learning.

    27. LF

      (laughs)

    28. VK

      But actually it was just iterative learning.

    29. LF

      Iterative learning.

    30. VK

      Um, and so this was there way back. I think what's interesting is, if, if you look at autonomous vehicles today, learning occurs, could occur in two pieces. One is perception. Understanding the world. Second is action, taking actions. Everything that I've seen that is successful is in the perception side of things. So in computer vision, we've made amazing strides in the last 10 years. So recognizing objects, actually detecting objects, classifying them, uh, uh, uh, and, and tagging them in some sense, annotating them. This is all done through machine learning. On the action side on the other hand, I don't know of any examples where there are fielded systems where we actually learn the right behavior.

  12. 33:52 – 37:04

    Perception limits: vision-only corner cases, LiDAR in mines, and energy costs

    1. LF

      So what do you think, just jumping out on topic since you mentioned autonomous vehicles, what do you think are the limits on the perception side? So I've talked to Elon Musk and there on the perception side, they're using primarily computer vision to perceive the environment. In your work with, because you work with the real world a lot and the physical world, what are the limits of computer vision? Do you think we can solve autonomous vehicles f- focusing, on the perception side, focusing on vision alone and machine learning?

    2. VK

      So, you know, we also have a spin-off company, Exyn Technologies, that, that, uh, works underground in mines.

    3. LF

      Wow.

    4. VK

      So you go into mines, they're, they're dark. They're dirty. You fly in a dirty area, there's stuff you kick up from, by the propellers, the downwash kicks up dust. I challenge you to get a computer vision algorithm to work there.

    5. LF

      Yeah.

    6. VK

      So we use LiDARS, uh, in that setting. Indoors and even outdoors when we fly through fields, I think there's a lot of potential for just solving the problem using computer vision alone. But I think the bigger question is can you actually solve or can you actually identify all the corner cases-

    7. LF

      Right.

    8. VK

      ... using a sense, single-sensing modality and using learning alone?

    9. LF

      So what's your intuition there?

    10. VK

      So, look, if you have a corner case and your algorithm doesn't work, your instinct is to go get data about the corner case and patch it up. Learn how to deal with that corner case. Uh, but at some point, this is gonna saturate. This approach is not viable.

    11. LF

      Mm-hmm.

    12. VK

      So today, computer vision algorithms can detect 90% of the objects or can detect objects 90% of the time, classify them 90% of the time. Um, cats on the internet, uh, I probably can do 95% on it.

    13. LF

      Okay.

    14. VK

      But to get from 90% to 99%, you need a lot more data. And then I tell you, "Well, that's not enough because I have a safety critical application. I wanna go from 99% to 99.9%." Well, that's even more data.

    15. LF

      Yeah.

    16. VK

      So I think if you look at wanting, uh, accuracy on the X-axis and look at the amount of data on the Y-axis, I believe that curve is an exponential curve.

    17. LF

      Wow. Okay. It's even hard if it's linear. (laughs)

    18. VK

      It's hard if it's linear, totally. But I think it's exponential.

    19. LF

      Okay.

    20. VK

      And the other thing you have to think about is that this process is a very, very power-hungry process. To run data farms-... or, or servers.

    21. LF

      Power, do you mean literally power? So energy.

    22. VK

      Literally power, literally power.

    23. LF

      Right.

    24. VK

      So n- in 2014, five years ago, and I don't have more recent data, 2% of US electricity consumption-

    25. LF

      (laughs)

    26. VK

      ... was from data farms.

    27. LF

      Wow.

    28. VK

      So we think about this as an information science and information processing problem. Actually, it is an energy processing problem. And so unless we figure out better ways of doing this, I don't think this is viable.

  13. 37:04 – 44:01

    Autonomous driving vs flight, and the future of delivery drones & flying cars

    1. LF

      So talking about driving, which is a safety-critical application, in some aspect, the flight is c- safety critical. Maybe philosophical question, maybe an engineering one, what problem do you think is harder to solve, autonomous driving or autonomous flight?

    2. VK

      That's a really interesting question. I think, um, autonomous flight has several advantages that autonomous driving doesn't have. So look, if I wanna go from point A to point B-

    3. LF

      Mm-hmm.

    4. VK

      ... um, I have a very, very safe trajectory, go vertically up to a maximum altitude, fly horizontally to, to just about the destination, and then come down vertically. Eh, this is pre-programmed. The equivalent of that is very hard to find in the self-driving car, car world because you're on the ground, you're in a two-dimensional surface, and the trajectories in the two-dimensional surface are more likely to encounter obstacles. Um, I mean this in an intuitive sense, but mathematically true, that's-

    5. LF

      Right.

    6. VK

      Mathematically as well, that's true. Uh-

    7. LF

      There's other option on the 2D space of platooning, or because there's so many obstacles, you can connect to those obstacles and all these kind of options.

    8. VK

      Sure, but those exist in the three-dimensional space as well.

    9. LF

      So they do. So, uh, d- the question also implies how difficult are obstacles in the three-dimensional space in flight?

    10. VK

      So, so that's the downside. I think in three-dimensional space, you're modeling three-dimensional world, um, not just, just because you wanna avoid it, but you wanna reason about it and you wanna work in that three-dimensional environment, and that's significantly harder. So tha- that's one disadvantage. I think the second disadvantage is, of course, anytime you fly, you have to put up with the peculiarities of aerodynamics. And they're complicated environments, how do you negotiate that? So that's always a problem.

    11. LF

      Do you see a time in the future where there is... You mentioned, um, uh, there's agriculture applications so there's a lot of applications of flying robots. But do you see a time in the future where there is tens of thousands or maybe hundreds of thousands of delivery drones that fill the sky? Uh, delivery, uh, flying robots?

    12. VK

      I think there's a lot of, uh, potential for the last mile delivery. And so in crowded cities, um, um, I don't know if you go... If you go to a place like Hong Kong, just crossing the river can take half an hour and while a drone can just do it in, in five minutes at most. I think you look at delivery of supplies to remote villages. I work with a nonprofit called WeRobotics. So they work in the Peruvian Amazon where the only highways are rivers and to get from point A to point B may take five hours, while with a drone, you can get there in 30 minutes. So just delivering drugs, retrieving samples for, for, for testing, uh, vaccines, uh, I think there's huge potential here. So I think it... The challenges are not technological. The, the challenge is economical. The one thing I'll tell you that nobody thinks about is the fact that we've not made huge strides in battery technology. Yes, it's true, batteries are becoming less expensive because we have these mega factories that are coming up, but they're all based on lithium-based technologies. And if you look at the energy density and the power density, those are two fundamentally limiting numbers. So power density is important because for a UAV to take off vertically into the air, which most drones do, they're not-

    13. LF

      Mm-hmm.

    14. VK

      ... they don't have a runway, um, you consume roughly 200 watts per kilo at the small size.

    15. LF

      Mm-hmm.

    16. VK

      That's a lot, right? In contrast, the human brain consumes less than 80 watts, the whole of the human brain. So just imagine-

    17. LF

      The-

    18. VK

      ... just lifting yourself into the air-

    19. LF

      (laughs)

    20. VK

      ... is like two or three light bulbs-

    21. LF

      Yeah.

    22. VK

      ... which makes no sense to me.

    23. LF

      Yeah, so you're going to have to, at scale, solve the energy problem then. Uh, charging the batteries, storing the, the energy, and so on. I think-

    24. VK

      And then the storage is the second problem. But storage limits the range.

    25. LF

      Right.

    26. VK

      But, uh, you know, you, you, you have to remember that you, you have to, you have to burn a lot of it for, for a given time, which is the-

    27. LF

      So the burning is another problem.

    28. VK

      ... which is the po- which is the power question.

    29. LF

      Yes. And do you think, just your intuition, there i- there are breakthroughs in batteries on the horizon? How hard is that problem?

    30. VK

      Mm... Look, there are a lot of companies that are promising flying cars-

  14. 44:01 – 52:18

    Human-robot collaboration, safety supervision, and the ethics of weaponization

    1. LF

      Yeah. That's the real challenge. Do you think it's possible for robots and humans to collaborate successfully on tasks? So a- a lot of robotics folks that I talk to and work with, I mean, humans just add a giant mess to the picture, so it's best to remove them from consideration when solving specific tasks. It's very difficult to model. There's just a source of uncertainty. In your work with, um, these agile, uh, flying robots, do you think there's a role for collaboration with humans, or is it best to model tasks in a way that, uh, that doesn't, uh, have a human in the picture?

    2. VK

      Well, I- I don't think we should ever think about robots without human in the picture. Ultimately, robots are there because we want them to solve problems for humans.

    3. LF

      Right.

    4. VK

      But there's no general solution to this problem. I think if you look at human interaction and how humans interact with robots-

    5. LF

      Mm-hmm.

    6. VK

      ... you know, we think of these in sort of three different ways. One is the human commanding the robot. The second is the human collaborating with the robot. So, for example, we work on how a robot can actually pick up things with a human and carry things.

    7. LF

      With a human, yeah.

    8. VK

      That's, like, true collaboration. And, uh, third, we think about humans as bystanders. Self-driving cars, what's the human's role and how does- how do self-driving cars acknowledge the presence of humans?

    9. LF

      Right.

    10. VK

      Um, so I think all of these things are different scenarios. It depends on what kind of humans, what kind of task, and I think it's very difficult to say that there's a general theory that we all have for this. But at the same time, it's also silly to say that we- we should think about robots independent of humans.

    11. LF

      Right.

    12. VK

      So, to me, human-robot interaction is almost a mandatory aspect of everything we do.

    13. LF

      Yes, so but to which degree? So what- your thoughts, if we jump to autonomous vehicles, for example, there's a- there's a big debate between what- what's called Level 2 and Level 4, so semi-autonomous and autonomous vehicles. And sort of the Tesla approach currently, at least, has a lot of collaboration between human and machine. So the human is supposed to actively supervise the operation of the robot. Part of the safety, uh, definition of how safe a- a robot is, in that case, is how effective is the human in monitoring it? Do you think that's ultimately not a good, uh, approach in sort of having a human in the picture, not as a bystander or, um, part of the infrastructure, but really as part of what's required to make the system safe?

    14. VK

      This is harder than it sounds.

    15. LF

      Right.

    16. VK

      I think, you know, if you- if you- if- I mean, I'm sure you've driven the- driven before on highways and so on. It's- it's really very hard to have to relinquish control to a machine and then take over when needed.

    17. LF

      Right.

    18. VK

      So I think Tesla's approach is interesting 'cause it allows you to periodically establish some kind of contact, uh, with the car. Toyota, on the other hand, is thinking about shared autonomy as a-

    19. LF

      Mm-hmm.

    20. VK

      ... or collaborative autonomy as a paradigm. If I may argue, these are very, very simple ways of human-robot collaboration 'cause the task is pretty boring. You sit in a vehicle. You go from point A to point B. I think the more interesting thing to me is, uh, for example, search and rescue. I've got a human-

    21. LF

      Right.

    22. VK

      ... first responder, robot first responders. I gotta do something. It's important. I have to do it in two minutes. The building is burning. There's been an explosion. It's collapsed. How do I do it? I think, to me, those are the interesting things where it's very, very unstructured, and what's the role of the human? What's the role of the robot? Uh, clearly, there's lots of interesting challenges, and as a field, I think we're gonna make a lot of progress in this area.

    23. LF

      Yeah, it's an exciting form of collaboration. You're right. In autonomous driving, the main enemy is just boredom of the human. (laughs)

    24. VK

      Yes.

    25. LF

      As- as opposed to in rescue operations, it's literally life and death, and, uh, the collaboration enables the effective completion of the mission, so it's exciting.

    26. VK

      Well, in some sense, you know, we're also doing this. You- you know, you think about the human driving a car, and almost invariably, the human's trying to estimate the state of the car, estimate the state of the environment, and so on. But what if the car were to estimate the state of the human?

    27. LF

      Mm-hmm.

    28. VK

      So, for example, I'm sure you have a smartphone, and the smartphone, uh, tries to figure out what you're doing and send you reminders.... and oftentimes telling you to drive to a certain place, although you have no intention of going there-

    29. LF

      (laughs)

    30. VK

      ... because it thinks that that's where you should be, uh, 'cause of some Gmail calendar entry, um, e- or, or, or something like that. And, and, you know, it's trying to f- constantly figure out who you are, what you're doing. If a car were to do that, maybe that would make the driver safer because the car is trying to figure out, is the driver paying attention, looking at his or her eyes, um-

  15. 52:18 – 56:31

    Big open problems and advice to students: breadth, math foundations, society

    1. LF

      And maybe we can elect some engineers to office as well on the other side. What are the biggest open problems in robotics? And you, you said we're in the early days in some sense. What are the problems we would like to solve in robotics?

    2. VK

      The... I think there are lots of problems, right? But I, I would phrase it, um, in the following way. Uh, i- if you look at the robots we're building, they're still very much tailored towards doing specific tasks in specific settings.

    3. LF

      Mm-hmm.

    4. VK

      I think the question of how do you get them to operate in much broader settings with where things can change in unstructured environments is up in the air. So, you know, think about self-driving cars. Uh, today, we can build a self-driving car in a parking lot. We can do level five autonomy in a, in a parking lot. But can you do level five autonomy in the streets of Napoli in Italy or Mumbai in India?

    5. LF

      Right. No.

    6. VK

      So in some sense, when we think about robotics, we have to think about where they're functioning, what kind of environment, what kind of a task. We have no understanding of how to put both those things together.

    7. LF

      So we're in the very early days of applying it to the physical world. And I was just in Naples actually and that's... there's levels of difficulty and complexity depending on which area you're applying it to.

    8. VK

      I think so. And we don't have a systematic way of understanding that.

    9. LF

      Right.

    10. VK

      You know, everybody says just 'cause a computer can now beat a human at any board game, we suddenly know-

    11. LF

      Yeah.

    12. VK

      ... something about intelligence. That's not true. A computer board game is very, very structured. It is the equivalent of working in a Henry Ford factory where things, parts come, you assemble, move on. It's a very, very, very structured setting. That's the easiest thing and we know how to do that.

    13. LF

      So you've done a lot of incredible work at, uh, the UPenn, University of Pennsylvania GRASP Lab. You're now Dean of Engineering at UPenn. What advice do you have for a new bright-eyed undergrad interested in robotics or AI or engineering?

    14. VK

      Well, I think there's really three things. One is, one is, uh, you, you have to get used to the idea that the world will not be the same in five years or four years whenever you graduate, right? Which is really hard to do. So this, this thing about predicting the future, every one of us needs to be trying to predict the future always.

    15. LF

      Mm-hmm.

    16. VK

      Um, not because you'll be any good at it, but by thinking about it, I think you sharpen your senses-

    17. LF

      Mm-hmm.

    18. VK

      ... and you become smarter. So that's number one. Number two, uh, and it's a corollary of the first piece, which is you really don't know what's gonna be important. So this idea that I'm gonna specialize in something which will allow me to go in a particular direction, it may be interesting, but it's important also to have this breadth so you have this jumping off point. Um, I think the third thing, and this is where I think Penn excels, I mean, we teach engineering, but it's always in the context of the liberal arts. It's always in the context of society. As engineers, we cannot afford to lose sight of that. So I think that's important. But I think one thing that people underestimate when they do robotics is the important of mathematical foundations-

    19. LF

      Mm-hmm.

    20. VK

      ... the important of represe- importance of representations. Not everything can just be solved by looking for ROS packages on the internet or to find a deep neural network that works. I think the representation question is key even to machine learning where if you hope, ever hope to achieve, uh, or get to explainable AI, somehow there need to be representations that you can understand.

    21. LF

      So if you wanna do robotics, you should also do mathematics and you said liberal arts, a little literature. If you wanna build a robot, you should be reading Dostoevsky. I agree with that.

    22. VK

      Very good. (laughs)

    23. LF

      (laughs) Vijay, thank you so much for talking today. It was an honor.

    24. VK

      Thank you. It was just a very exciting conversation. Thank you.

Episode duration: 56:46

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Transcript of episode HYsLTNXMl1Q

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

iOSAndroidClaudeChrome