Now, a professor at Oregon State University is working to instill in robots the movement capabilities of animals and humans. I spoke recently with Jonathan Hurst about the future of running and walking robots.
There are already robots that can walk and run. How will your technology be different?
It's getting to be a little bit of a crowded field, which is great. But nobody has even approached what animals are capable of doing. We see some robots, which work very well in a controlled environment, walking on very flat ground. It looks very fluid, but it doesn't work well if it hits any disturbances, any bump in the ground. [There] are completely passive walkers, robots that aren't really robots. They're just mechanical pieces. They don't have any electronics on them at all. You start them walking down a hill and they look very natural, but they're very sensitive to disturbances. Another example would be BigDog, which is at Boston Dynamics. This is a military project to basically build a robotic packing mule, something that can assist soldiers in carrying heavy loads. This robot burns a huge amount of energy. All of its control is active. It's analogous to getting around in the world by squatting and jumping and stopping over and over. It uses a lot of energy. What we're trying to do is combine the idea of the passive dynamic walkers -- which don't have any motors and are very efficient but sensitive to disturbances -- with the robustness and ability to deal with disturbances of robots like BigDog.
Why do we want robots to move more like humans and animals?
There are a number of reasons. If we can really understand how animals do it, that means we can replicate it in a robot. If we've demonstrated in a robot human- or animal-like locomotion, that means we can build exoskeletons and prosthetic limbs for humans. I expect to see this sort of thing coming out very soon, the next 10 years. We want robots getting around in human environments. We're going to want robots in our homes. We're going to want robots in hospitals, taking care of people. We're also going to want robots in construction sites. We're going to want robots in military situations to get around rough urban terrain, up and down stairs, through doors.
Why is it so difficult to replicate human and animal walking and running in robots?
I might ask, "Why do people think it's so easy?" But walking and running is a very complicated dynamic motion. If you look at a grandfather clock, it took hundreds of years to develop. Walking and running is a lot more complicated than that. The only reason we know anywhere near as much as we do about it is we can look at animals and try to get some inspiration and understanding from that.
My work is on the science side of things and a little bit removed from just engineering. I'm trying to really understand how animals work and how we can optimize machines. It starts [with] biologists who study animals and study the disturbances and responses of animals. If an animal is walking along and they hit a sudden drop in the ground surface or a change of the ground's stiffness, they'll make some observations of what the animal does. You can only guess really at what's going on inside the animal. It's this tightly coupled combination of neural control with muscle activation and the passive dynamics of springing tendons and things like that. You come up with hypotheses and theories and you try to test them. The way that we're trying to confirm all that is by replicating it with a robot. If we can show that the robot exhibits the same behavior as the animal when faced with the same disturbance, it's not a proof but it's a pretty strong hint that we have an understanding of what's going on. My focus is on creating a mechanical system that enables complex physical interaction tasks like walking and running.
What do you want your robot to be able to do?
I'm hoping to achieve a robot that under normal circumstances on flat ground, it uses almost entirely passive dynamics. So it just bounces along. It's got the springs in the right place so the leg swings along and it doesn't really take any work from motors. And as soon as you hit some unexpected disturbance like a sudden drop in the ground surface, that's when the software can take over, the motors can do some work, and get you back into your passive cycle. By doing this, we can take very little energy. The mechanical system becomes just this cycle. If you think of it as a pogo stick, you're just bouncing along. That's different from the squat, jump, wait. I'm trying to incorporate these energy systems in the mechanical systems as part of the passive dynamics.
How far along are you in this process?
This summer we're building our first prototype monopod. It's just one leg. It will be planar. It's on a boom. It can't fall side-to-side, it can only fall forward and backward. We'll be testing that simple system right after we build it. We'll be adding an additional leg to it, so it will be a biped. We will be comparing the behavior of this biped directly to a running ostrich. We'll have a sudden drop in the ground surface. We'll measure the forces on the ground and so on for both the robot and the ostrich. We hope to learn something scientific about how animals run and they respond to these disturbances. That is on a three-year grant. My long-term goal is to take these off the boom and to do 3-D running out in the world, but that's a little bit less predictable.
Image, top: Biped robot / Courtesy of Oregon State University
Image, bottom: Jonathan Hurst