Subscribe Now on Apple Podcasts, Spotify, and Stitcher!

For robots to be more useful around people, they’ll need to go where we go. But how? Associate Professor Jonathan Hurst thinks the answer is simple. Walking. But actually making a walking robot is no simple feat.

Jonathan Hurst

Jonathan Hurst, associate professor of mechanical engineering and robotics, with Cassie, a bipedal robot created in his lab.

Bonus Content

Transcript

[Clip from Star Wars Episode IV: A New Hope. Twentieth Century Fox, 1977.]

THREEPIO: How did I get into this mess? I really don't know how. We seem to be made to suffer. It's our lot in life.

ARTOO: [beeping sounds]

THREEPIO: I've got to rest before I fall apart. My joints are almost frozen.

ARTOO: [beeping sounds]

PERRY: It’s an iconic shot in a classic movie. Two robots – or droids to be precise – walking across the barren wastes of the desert planet Tatooine.

THREEPIO: What a desolate place this is. Where are you going? Well, I'm not going that way. It's much too rocky. This way is much easier.

When you think about it, for all of C3PO’s whining, the fact that he’s actually walking is pretty incredible. Or it would be if Star Wars wasn’t a work of fiction. (Sorry fellow nerds.)

Fiction or not, it’s important to note just how much pop culture shapes how we think about pretty complex things, especially robots. From Metropolis to The Terminator, we see humanoid robots walking along side real humans like it’s no big deal.

So how come we don’t have robots walking around everywhere?

Well, it turns out that making a robot walk is actually kind of hard. But here at Oregon State, we’re getting pretty good at it.

[MUSIC: The Ether Bunny, by Eyes Closed Audio, used with permission under a Creative Commons Attribution License]

From the College of Engineering at Oregon State University, this is “Engineering Out Loud.”

I’m your host, Owen Perry, and this season of Engineering Out Loud is all about robots and artificial intelligence and where those technologies are going. How close are we from the stuff of science fiction becoming real life?

[MUSIC: Finding Movement by Kevin MacLeod, part of the YouTube Audio Library. Licensed under a Creative Commons license]

If you talk to a lot of roboticists, which I happen to do, it’s not hard to imagine a day in the very near future when it will seem perfectly normal to pass your local delivery bot strolling down the sidewalk.

In this episode we’re going to look at a very literal part of that journey: How will robots even get around? But first, why would we even want them to?

HURST: Why do we care about robots walking? There are a couple of different answers to that. Having legs that are as capable as human legs are going to enable robots to work with us around us in human environments and really be a valuable tool, an important tool for us. It's the pragmatic of having robots in the world and in our environment and in all the places we can go and all the places we can't go, like burning buildings. And it's the scientific interest of understanding how something works and understanding our world around us.

PERRY: That was Jonathan Hurst, an associate professor of robotics and the chief technology officer at Agility Robotics, a company that spun out of his research at Oregon State.

HURST: My goal is to understand legged locomotion, to effectively be able to write down the answer and say these are the key parts that make legged locomotion work.

PERRY: Yes, legged locomotion. AKA walking. It’s how we get around, so if we want robots to be helpful, they will need to walk with us.

HURST: Clearly, we've built our environment around us and our morphology, and we can get around it quite well, and people who are disabled who do not have the use of their legs have great difficulty in getting around the same environment, as do robots. If you have wheels or tracks on a robot, they often are stuck at the curbs or the stairs and so on. Perhaps there are other great solutions to get around in our environments, but we know that legs are one of them.

PERRY: Let’s take a moment to consider the wonder that is the human gait.

[MUSIC: Hammock Fight by Kevin MacLeod, part of the YouTube Audio Library. Licensed under a Creative Commons license]

PERRY: Wikipedia tells us that “human gait is defined as bipedal, biphasic forward propulsion of the center of gravity of the human body, in which there are alternate sinuous movements of different segments of the body with least expenditure of energy.”

[SOUND EFFECT: Record scratch]

Okay, let’s take a minute to think about what it takes to actually walk.

[MUSIC: The Curious Kitten by Aaron Kenny, part of the YouTube Audio Library. Licensed under a Creative Commons license]

I wake up in the morning — usually to the sound of my crying one-year-old. I throw one leg out from under the covers and onto the floor followed by the other, then stand up and proceed to move through the darkened room, nerves in my feet sending signals to my brain telling it to hurry up and put socks on because this floor is cold. I get to the end of the hall, step over the stuffed animal left in the middle of the floor, and — ouch — directly on the Lego piece. And eventually, I make it to my son’s crib, pick him up, pivot, and bring him over to the changing table. That’s just the first 45 seconds of my day.

How do your muscles and bones and brain work together to get you where you want to go? Now, how do you make a robot do that? And once you have a robot that can theoretically walk, how do you even start teaching it to do so?

HURST: It's an incredibly dynamic behavior, and we just don't fully understand it yet. So it's a science problem. I want to understand it.That's what science is about.

PERRY: Jonathan is not trying to find the best way to replicate a human though. He’s trying to break down the basic understanding of how legged locomotion works and then apply that knowledge to the design of the robot.

HURST: And that's a very different approach from saying, “We're going to build a robot that looks physically like a human or it looks physically like some other kind of animal.” We're not copying the locations of the joints or where the muscles are, where the springs are. We're trying to understand how and why it works first, then do the engineering to implement that.

PERRY: He looked at representations that other researchers have observed or hypothesized in studies of animals to mathematically explain their locomotion.

HURST: Our robots are founded on our best hypotheses about our understanding of the dynamics behind all animal gaits. One of these is the spring mass model, and it's basically just a point mass in a massless spring for the leg and two of these springs to make a bipedal robot.

PERRY: Think of a child on a pogostick. The child provides the mass pressing down on the spring. On each bounce, the spring is compressed and then expands, lifting itself (and the kid) off the ground using relatively little energy. Using this springy model, robots could theoretically walk and run with remarkable energy economy.

HURST: And what's really interesting about this extremely simple math model is that it's actually fairly complex to simulate in how it's incredibly non linear, and it's an incredibly good representation for all animals that walk and run — ghost crabs to horses, to humans to ostriches — the ground reaction forces that you get, the center of mass motion that you get all matches this really simple model in the steady state, continuous walking, continuous running behaviors.

And then we build a robot as an engineer would from engineering principles to try and create that certain set of dynamics.

PERRY: The first robot Jonathan built using this model was named ATRIAS.

HURST: ATRIAS is the first robot in history to walk like a human.

[MUSIC: Dreams Electric by Geography, part of the YouTube Audio Library. Licensed under a Creative Commons license]

PERRY: ATRIAS, that’s A-T-R-I-A-S. It’s an acronym for Assume the Robot is a Sphere. Assuming something is a sphere is an old physicist's joke about simplifying the properties of an object you’re studying until they're easy to represent with math, like a sphere.

Based on a the spring-mass model and the combined passive dynamics of a mechanical system with computer control, ATRIAS has the ability to react to rough terrain, maintain balance, retain an efficiency of motion, and essentially walk like humans do.

The system is also efficient. Studies done by Jonathan’s group with their ATRIAS robot model, which incorporates the spring-mass theory, showed that it's three times more energy-efficient than any other human-sized bipedal robots.

But the design wasn’t perfect.

HURST: ATRIAS I mentioned has those two motors right at the hip and then that four bar linkage for the legs. Well, it turns out that that configuration has a problem that hadn't really been examined closely in robotics before. We dubbed it antagonistic work where for the particular task we're trying to do, both of the motors have to apply torque in equal and opposite directions to apply force to the ground, but they are going to move in the same direction to swing leg backwards. And so that means that one motor is outputting a lot of very positive power and the other motor is absorbing a lot of power, doing a lot of negative work, and it's this internal kind of power loop between those two motors that's just internal to the machine. One motor is acting like a brake all the time, dragging things down while the other motor has to overcome that brake and do all the work in the world that is needed to keep the robot walking or running.

PERRY: Luckily for Hurst, the motors were spec’d to be more than twice as big as they needed to be in case he ran into something like this.

So while ATRIAS’s design wasn’t perfect, it confirmed Hurst’s confidence in using a spring-mass model as the way to go for building a walking robot. He and his research group set out to optimize the design principles of ATRIAS, capturing animal-like dynamics while producing a robot more capable of standing on its own. What they came up with is Cassie, a bipedal robot that looks like, well, it looks like a pair of legs, but maybe not the kind of legs you’re thinking of.

HURST: The best analog would be, it looks like an ostrich leg, you know, with a very short thigh up at the top. And then long links the rest of the way down, standing on little toes. But we didn't set out to make it look like an ostrich. We set out to find a leg configuration that still has the compliance where we want it and avoids this antagonistic work loop, which is a function of where the actuators are located. What's exciting to me is that we end up with something that looks like some animal. Maybe we're actually getting at some of the reasons why an animal is shaped that way. But any similarity to an animal that is generated by our robot is, is for that kind of reason, for some sort of basic engineering or science reason.

PERRY: If you think about it, that makes a lot of sense. An ostrich can run very fast, clocking in at up to 43 miles an hour, but it also runs very efficiently.

The team took what they learned from ATRIAS and applied it to Cassie with a number of improvements.

HURST: ATRIAS can only walk and run and nothing else. It can't steer and can't even stand in place. So the next step with Cassie was adding a few more motors to it. Instead of six motors total, there's 10. Cassie can stand in place because it's got ankles.Cassie has a yaw or direction on the, on the hips so we can steer. And Cassie was a great big step up in engineering capability. The robot is half as heavy. Some all new transmissions that we designed that really improved performance quite a lot.

PERRY: Cassie then became the basis for spinning off Agility Robotics, which to date has built and sold more than 10 of the robots to research programs at other universities around the world.

HURST: This is the only bipedal machine anybody can buy anywhere that is really dynamically capable. That is, for which it is even possible to do this kind of spring mass gait, this kind of spring mass behavior. So if you don't have the joints in the right place, if you don't have the dynamics of the hardware, it doesn't matter what software you write, it's not, you can't control a machine to do just anything. It has to actually be physically capable of doing it, and Cassie can.

Part of the exercise and starting a company is finding the best use case, finding the use case that people are willing to pay money for.

PERRY: One of those use cases is package delivery. Distribution centers are becoming more and more automated, and the reality of driverless delivery trucks is on the horizon.

HURST: Now you're seeing a whole logistics chain that's automated from a manufacturer all the way to your curb. But then that last piece, they're trapped.

PERRY: Without a person, how does your package get from the truck to your doorstep? It’s a conundrum that a lot of companies are looking into.

HURST: You know, there's wheeled vehicles, there's the flying drones. Those will have their place. Those are gonna be successful. Those are gonna be part of the solution. But there's gonna be a lot of places where you're not, they won't get, and our robots will with the legs.

PERRY: Since we recorded this interview, Agility has announced a partnership with Ford Motor Company to explore equipping their autonomous delivery vehicles with robots that can be deployed to walk the final steps to your door.

HURST: In the past ten, 15 years, our moment of going from, “Legged locomotion is magical and we don't understand it at all,” to, “Hey, we actually kind of know what's going on here.”

[MUSIC: Dreams Electric by Geography, part of the YouTube Audio Library. Licensed under a Creative Commons license]

HURST: I think we're past the point now where, a research lab can say, “Look, I have successfully made this robot that can place one foot in front of the other.” That's no longer considered a success. We've moved beyond that.

PERRY: One of those ways Jonathan and Agility are moving beyond is Digit, a robot that adds a torso and arms on top of Cassie’s legs, and it even adds little headlike sensor array.

HURST: Our kind of vision for this potential use case is logistics and package delivery, and we imagine having, these robots, these Digit robots with arms and perception, kind of as the last piece of the delivery coming out of an autonomous vehicle and carrying a package to your doorstep.

PERRY: But arms are useful in other ways too.

HURST: You might notice if you're trying to walk or run outside, if you cross your arms around your chest and you try to walk or run, you have to twist your shoulders a lot. You really can't control your shoulder orientation cause you swing your leg forward, and you have to twist your body the other way. The inertia is what it is. We use our arms to counter a lot of that. We've added a pair of arms and I would call them more like bilaterally, symmetrical tails because they're really there for that balancing of the yaw, and other degrees of freedom. So you can swing it around. Also the arms are gonna be there to catch the robot when it falls to reorient the body to pick it back up and so on, and we're starting to add perception and sensors too.

The next step is going to be doing the,building an organization, building the investment to actually make it reliable and functional in the world.

PERRY: So how long will it be before robots like Digit are walking down the street, as ubiquitous as cars are today?

HURST: Well, we actually look at the history of the automobile kind of as an example for that. The automobile industry grew, it's something like 50% a year between 50% and 100% a year for, I mean, a long time, a hundred years, something like that. And so early on they were really fairly specialized industrial use cases. And then they were really interesting curiosities for awhile that a few people owned for whatever purposes they needed and gradually became much more capable and much more common. And I think that'll be the same thing with the legged robots out in the world. I think it's going to be a very close parallel. I think it's going to be an industry similar size, having robots with legs. We're going to have as many of those as we have cars now.

[MUSIC: Finding Movement by Kevin MacLeod, part of the YouTube Audio Library. Licensed under a Creative Commons license]

HURST: We've moved legged locomotion from being a science problem that we don't quite understand to an engineering problem where we just need to make better engineered systems with springs here and transmissions there and motors there and improve our electronics here. But we basically know the physics of how it works. That's, that's what's exciting to me. That's what motivates my research.

PERRY: Robots walking among us ...why does that seem a little scary?

HURST: I think the reason people get concerned about it is because robots are visible. They are anthropomorphic. They kind of look maybe like a person, and they are in our space. That's the only thing that's different.

PERRY: And maybe Hollywood has something to do with it?

[Clip from The Terminator, Hemdale, 1984]

Kyle Reese: Listen, and understand! That Terminator is out there! It can't be bargained with. It can't be reasoned with. It doesn't feel pity, or remorse, or fear. And it absolutely will not stop... ever, until you are dead!

PERRY: Yeah, that might be it. Here’s Hurst’s take on the impact of the Terminator.

HURST: That and some other movies are the first experienced a lot of people have had with robots and robotics. And so, that's why those fears come out is because of a relatively uninformed narrative that is the first one that most people saw.

PERRY: But there’s something more to it than that.

HURST: I think people assume that these robots have a lot more agency than they do. In other words, people assume when they see our robot walking in a way that looks somewhat natural. They then also assume that that the robot can talk to them or can make decisions about where it's going or any number of things that that comes from anthropomorphizing the machine when you see it do one thing capably.

PERRY: That brings up another concern that Hurst often hears, that robots are coming for our jobs.

HURST: You’ve got to think of these robots simply as fairly sophisticated tools for humans to be able to do something better than they did before. And I think that's what's really important to remember, is that we want to do more. We don't want to just do what we're doing now and then have a whole bunch of people unemployed. We want everybody to be employed, but we want to do more. I think the problem comes in when something changes too fast. You know, if next year you could just buy a widget that you put in your in your vehicle or a tractor trailer and all of a sudden it's now an autonomous vehicle. That's a lot of people who are out of work, taxi drivers and truck drivers. I don't think that's what's going to happen. I don’t know; it's going to be a challenge for us, and we're going to need to change.

[MUSIC: Dreams Electric by Geography, part of the YouTube Audio Library. Licensed under a Creative Commons license]

PERRY: This episode was produced and hosted by me, Owen Perry. Audio editing was by Molly Aton and production assistance by Rachel Robertson. Our intro music is “The Ether Bunny” by Eyes Closed Audio. You can find them on SoundCloud and we used their song with permission of a Creative Commons attribution license. Other music and sound effects in this episode were also used with appropriate licenses.

For more episodes and bonus content, visit engineeringoutloud.oregonstate.edu. Also, search for “Engineering Out Loud” on your favorite podcast app, and do us a solid, please subscribe.

It feels like we need to go out on a bit of levity. Hey, Jon the Robot, what do you got?

JON THE ROBOT: Have you heard that Cassie is a super efficient bipedal robot? It walks all over the competition.

The competitors don't have a leg to stand on.

PERRY: Right. What else you got?

JON THE ROBOT: Cassie literally runs circles around the competition. The other robots used to hold up, but nowadays they've really lost their footing.

Doing this was no small feat... it required a dynamically consistent set of torques, ground reaction forces, and generalized accelerations that minimize errors in the operational space accelerations.

Do you get it? Feet.

PERRY: Okay. Thanks Jon.

Jon the Robot was created by Naomi Fitter, an assistant professor or robotics here at Oregon State, who wrote those jokes along with her husband John Vilk. We’ll hear from Naomi in a later episode this season. Until then, catch you later.