Engineers have designed robots that crawl, swim, fly and even slither like a snake, but no robot can hold a candle to a squirrel, which can parkour through a thicket of branches, leap across perilous gaps and execute pinpoint landings on the flimsiest of branches.
University of California, Berkeley, biologists and engineers are trying to remedy that situation. Based on studies of the biomechanics of squirrel leaps and landings, they have designed a hopping robot that can stick a landing on a narrow perch.
The feat, to be reported in the March 19 issue of the journal Science Robotics, is a big step in the design of more agile robots, ones that can leap among the trusses and girders of buildings under construction or robots that can monitor the environment in tangled forests or tree canopies.
"The robots we have now are OK, but how do you take it to the next level? How do you get robots to navigate a challenging environment in a disaster where you have pipes and beams and wires? Squirrels could do that, no problem. Robots can't do that," said Robert Full, one of paper's senior authors and a professor of integrative biology at UC Berkeley.
"Squirrels are nature's best athletes," Full added. "The way that they can maneuver and escape is unbelievable. The idea is to try to define the control strategies that give the animals a wide range of behavioral options to perform extraordinary feats and use that information to build more agile robots."
Justin Yim, a former UC Berkeley graduate student and co-first author of the paper, translated what Full and his biology students discovered in squirrels to Salto, a one-legged robot developed at UC Berkeley in 2016 that could already hop and parkour and stick a landing, but only on flat ground. The challenge was to stick the landing while hitting a specific point — a narrow rod.
"If you think about trying to jump to a point — maybe you're doing something like playing hopscotch and you want to land your feet in a certain spot — you want to stick that landing and not take a step," explained Yim, now an assistant professor of mechanical science and engineering at the University of Illinois, Urbana Champaign (UIUC). "If you feel like you're going to fall over forward, then you might pinwheel your arms, but you'll also probably stand up straight in order to keep yourself from falling over. If it feels like you're falling backward and you might have to sit down because you're not going to be able to quite make it, you might pinwheel your arms backward, but you're likely also to crouch down as you do this. That is the same behavior that we programmed into the robot. If it's going to be swinging under, it should crouch. If it's going to swing over, it should extend out and stand tall."
Using these strategies, Yim is embarking on a NASA-funded project to design a small, one-legged robot that could explore Enceladus, a moon of Saturn, where the gravity is one-eightieth that of Earth, and a single hop could carry the robot the length of a football field.
The new robot design is based on a biomechanical analysis of squirrel landings detailed in a paper accepted for publication in the Journal of Experimental Biology and posted online Feb. 27. Full is senior author and former graduate student Sebastian Lee is first author of that paper.
Mixing biology and robotics
Salto, short for Saltatorial Agile Locomotion on Terrain Obstacles, originated a decade ago in the lab of Ronald Fearing, now a Professor in the Graduate School in UC Berkeley's Department of Electrical Engineering and Computer Sciences (EECS). Much of its hopping, parkouring and landing ability is a result of a long-standing interdisciplinary collaboration between biology students in Full's Polypedal Lab and engineering students in Fearing's Biomimetic Millisystems Lab .
During the five years Yim was a UC Berkeley graduate student — he got his Ph.D. in EECS in 2020, with Fearing as his adviser — he met with Full's group every other week to learn from their biology experiments. Yim was trying to leverage Salto's ability to land upright on a flat spot, even outdoors, to get it to hit a specific target, like a branch. Salto already had a motorized flywheel, or reaction wheel, to help it balance, much the way humans wheel their arms to restore balance. But that wasn't sufficient for it to stick a direct landing on a precarious perch. He decided to try reversing the motors that launch Salto and use them to brake when landing.
Suspecting that squirrels did the same with their legs when landing, the biology and robotics teams worked in parallel to confirm this and show that it would help Salto stick a landing. Full's team instrumented a branch with sensors that measured the force perpendicular to the branch when a squirrel landed and the torque or turning force with respect to the branch that the squirrel applied with its feet.
The research team found, based on high-speed video and sensor measurements, that when squirrels land after a heroic leap, they basically do a handstand on the branch, directing the force of landing through their shoulder joint so as to stress the joint as little as possible. Using pads on their feet, they then grasp the branch and twist to overcome whatever excess torque threatens to send them over or under the branch.
"Almost all of the energy — 86% of the kinetic energy — was absorbed by the front legs," he said. "They're really doing front handstands onto the branch, and then the rest of it follows. Then their feet generate a pull-up torque, if they're going under; if they are going to go over the top — they're overshooting, potentially — they generate a braking torque."
Perhaps more important to balancing, however, they found that squirrels also adjust the braking force applied to the branch when landing to compensate for over- or undershooting.
"If you're going to undershoot, what you can do is generate less leg-breaking force; your leg will collapse some, and then your inertia is going to be less, and that will swing you back up to correct," Full said. "Whereas if you are overshooting, you want to do the opposite — you want to have your legs generate more breaking force so that you have a bigger inertia and it slows you down so that you can have a balanced landing."
Yim and UC Berkeley undergraduate Eric Wang redesigned Salto to incorporate adjustable leg forces, supplementing the torque of the reaction wheel. With these modifications, Salto was able to jump onto a branch and balance a handful of times, despite the fact that it had no ability to grip with its feet, Yim said.
"We decided to take the most difficult path and give the robot no ability to apply any torque on the branch with its feet. We specifically designed a passive gripper that even had very low friction to minimize that torque," Yim said. "In future work, I think it would be interesting to explore other more capable grippers that could drastically expand the robot's ability to control the torque it applies to the branch and expand its ability to land. Maybe not just on branches, but on complex flat ground, too."
In parallel, Full is now investigating the importance of the torque applied by the squirrel's foot upon landing. Unlike monkeys, squirrels do not have a usable thumb that allows a prehensile grasp, so they must palm a branch, he said. But that may be an advantage.
"If you're a squirrel being chased by a predator, like a hawk or another squirrel, you want to have a sufficiently stable grasp, where you can parkour off a branch quickly, but not too firm a grasp," he said. "They don't have to worry about letting go, they just bounce off."
One-legged robots may sound impractical, given the potential for falling over when standing still. But Yim says that for jumping really high, one leg is the way to go.
"One leg is the best number for jumping; you can put the most power into that one leg if you don't distribute that power among multiple different devices. And the drawbacks you get from having only one leg lessen as you jump higher," Yim said. "When you jump many, many times the height of your legs, there's only one gait, and that is the gait in which every leg touches the ground at the same time and every leg leaves the ground at approximately the same time. So at that point, having multiple legs is kind of like having one leg. You might as well just use the one."
Other co-authors of the Science Robotics paper are Fearing and former UC Berkeley undergraduate Eric Wang, now a graduate student at MIT, and former graduate student Nathaniel Hunt, now an associate professor at the University of Nebraska in Omaha. Co-authors of the J. Exp. Bio. paper are Wang, Hunt, Fearing, UC Berkeley Associate Professor of Mechanical Engineering Hannah Stuart and former UC Berkeley undergraduates Stanley Wang and Duyi Kuang. The research was funded by the U.S. Army Research Office (W911NF-18-1-0038, W911NF-1810327) and the National Institutes of Health (P20GM109090).