Yim's jumping robot inspired by squirrels' branch-to-branch leaping technique

Mechanical science and engineering assistant professor Justin Yim’s research on jumping robots that leverage the biomechanical capabilities of squirrels was recently published in the high-impact journal Science Robotics. The study, “Monopedal robot branch-to-branch leaping and landing inspired by squirrel balance control,” was a product of his doctoral research and was done in collaboration with his advisor, Professor Ronald Fearing, committee member Professor Robert Full, fellow graduate students (now Drs.) Sebastian Lee and Nathaniel Hunt (now at the University of Nebraska Omaha), and undergraduate researcher Eric Wang (now a PhD candidate at Massachusetts Institute of Technology)—all from his alma mater, the University of California, Berkeley.

While biologists at Berkeley were already studying squirrels’ biomechanics, Yim found opportunity to learn from them in a new way—to mimic their behaviors in jumping robots. “Being able to jump between narrow branches is something that could enable jumping robots to operate in environments that are challenging for humans and other types of robots,” he said.

While squirrels commonly jump from branch to branch, what may be less obvious to the observer is the fact that they often tend to slightly undershoot or overshoot landings due to the challenges associated with navigating sparse environments. Nevertheless, they very rarely fall—an extremely desirable ability to replicate in robots. Squirrels’ ability to stay upright may be attributed to a number of biological characteristics such as their fault-tolerant, fail-safe musculoskeletal system. However, their athletic performance is also remarkable, with a tendency to consistently produce upright landings that inherently prepare them to execute the next leap.

The forces applied by a foot during locomotion are described by four classifications of control strategies: vertical, horizontal (i.e., forward/backward), torque, and footstep placement. This means that for each footstep, there will be force controlling how hard the foot comes down to the ground, how much the foot pushes forward or backward, how much ground it covers during the step, and how much torque (i.e., twist) it applies to the ground from heel to toe. Torque and footstep placement are helpful in remaining upright when there is solid ground beneath the foot, but these strategies no longer help when the ground surface is a narrow beam. In this circumstance, vertical and horizontal forces become more prevalent.

“Reaction wheels can actuate that forward or backward force,” said Yim, noting that implementing reaction wheels, which is akin to wheeling one’s arms or bending one’s body while balancing on a tightrope, is a go-to strategy for balance in robots. “However, that’s not the only force we can apply. Like the squirrels, we can make the robot stand up tall or crouch down as it lands. Intuitively, it may seem like that wouldn’t do much to help balance, but our mathematical models showed that it actually makes a really big difference.”

Coauthor Lee collected data on the squirrels’ biomechanics by enticing them to participate in homemade, IRB-approved parkour courses and capturing their motion over the obstacles with a highspeed camera.

“Like the U of I, Berkeley has a ton of squirrels running around,” Yim said, explaining that the squirrels tended to congregate regularly in a particular eucalyptus grove where the courses were eventually set up. He then translated findings from the squirrels’ movements into an iterated version of his jumping robot, Salto.

“We could tell the robot to stand up taller or crouch down faster for better [jump] performance,” he said.

Yim’s exploration of jumping robot technology has since continued through new projects such as LEAP, a concept for jumping on Saturn’s moon Enceladus.

“I think the biggest impact here is exploring how we can make big motions in constrained environments,” he said. “[For applications like] environmental monitoring, firefighting, and forestry management, as well as inspection or maintenance of industrial environments that have things like trusses, pipes, or girders, having a robot that could navigate narrow beams opens up new avenues where robots currently can’t move at all.”