The UC Berkeley Hybrid Robotics group has demonstrated the first experimental instance of dynamic walking on stepping stones by a bipedal robot, according to Koushil Sreenath, the campus assistant professor of mechanical engineering who leads the group.
After more than three years of research, the Hybrid Robotics group, in partnership with researchers at Carnegie Mellon University, has developed algorithms that allow the ATRIAS robot to scale “discrete terrain” where there may be gaps between footholds and differences in foothold heights, according to Sreenath.
The ability to traverse such terrain makes legged robots more versatile than their wheeled counterparts, according to Ayush Agrawal, a campus mechanical engineering doctoral student and member of the Hybrid Robotics group. Sreenath compared designing bipedal robots to self-driving cars, saying that the varied terrain creates more challenges.
“It’s a lot more challenging problem, because you no longer have a flat ground, you no longer have lane markings, every time you step on something … the terrain can move,” Sreenath said.
The bipedal ATRIAS robot used by the group utilizes dynamic walking. Because it lacks “feet,” it cannot hold its position by staying still but must step in place, much like wearing stilts, according to Sreenath.
The group will soon be working with a newer robot called Cassie, whose “feet” and five leg motors — as opposed to ATRIAS’s three — will provide greater control as well as greater complexity in movement, according to Agrawal.
With the advancement of the algorithms used with ATRIAS and other legged robots, the technology could be also used to create “smart” prostheses that help humans decide where to step, according to Sreenath. He added that the Hybrid Robotics group is currently partnered with the French startup Wandercraft, which is developing exoskeletons for walking-impaired users, and implementing autonomy could be the next step.
Legged robots could also have applications in emergency situations and space exploration, Agrawal said. Sreenath added that while it is difficult to predict when such technology will become available, he said he believes it will be relatively soon.
“It’s hard to tell. … It could be five years, it could be fifteen years — definitely within our lifetimes,” said Sreenath.