Researchers from Carnegie Mellon University, the University of California, Berkeley, and Facebook AI have taught a four-legged robot to adapt to different conditions while walking – in real-time. The robot was able to walk over rocks, mud, sand, stairs, slippery surface, through a construction site, and around outdoor terrain.
A major obstacle to deploying legged robots – whether with two, four, or even more legs – is figuring out how the robot will respond to changing conditions. Humans can adapt to different conditions and adjust to carrying a heavy backpack or limp along with an injured ankle. But, legged robots cannot adjust so quickly.
Seeking to change that, the research team has developed a Rapid Motor Adaptation (RMA) algorithm that helps four-legged robots adjust in real-time to unseen scenarios such as changing terrains, changing payloads, wear and tear.
RMA consists of two components: a base policy and an adaptation module. The combination of these components enables the robot to adapt to novel situations in fractions of a second.
The team tested RMA by deploying it on the Unitree A1 robot without any fine-tuning. The algorithm does not rely on any hand-coded motions and allows the robot to learn how to adapt quickly through trial and error and interact with the surroundings. The robot learns from how its body reacts on different surfaces, just the way humans do.
“The focus is not walking. It is learning,” said Pathak, an assistant professor in the Robotics Institute at CMU. “By falling thousands of times or millions of times in simulation, it learns to walk from scratch and adapts to the ever-changing real world. Since the algorithm’s focus is learning, it is applicable to any kind of robot, not just this one.“
Testing showed that the RMA-enabled robot outperformed competing systems when walking over varied surfaces, slopes, and obstacles and when carrying different payloads.
“If you pick up a backpack, you adjust your motion without knowing the exact weight. If the terrain beneath your feet changes, you adjust your balance to compensate. RMA does this by adapting the robot joints in real-time,” said Ashish Kumar, a Ph.D. student at Berkeley.