Fall 2021 GRASP SFI: Ashish Kumar, UC Berkeley, “Rapid Motor Adaptation for Legged Robots”

Legged locomotion is commonly studied and expressed as a discrete set of gait patterns, like walk, trot and gallop, which are usually treated as given and pre-programmed in legged robots. However this limits the generality of locomotion. To achieve this general locomotion capability, I will talk about two important capabilities we add to our A1 robot – fast online adaptation and emergent gaits. To adapt in real-time to unseen scenarios like changing terrains, changing payloads, wear and tear I will present the Rapid Motor Adaptation (RMA) algorithm. RMA enables the robot to adapt to novel situations in fractions of a second. RMA is trained completely in simulation without using any domain knowledge like reference trajectories or predefined foot trajectory generators and is deployed on the A1 robot without any fine-tuning. We deploy it on a variety of difficult terrains including rocky, slippery, deformable surfaces in environments with grass, long vegetation, concrete, pebbles, stairs, sand, etc. I will further demonstrate that the same learning framework shows emergent gaits on simple terrains when we give different target speeds to the algorithm. This concurs with the recent animal motor studies which show that conventional gaits are prevalent in ideal flat terrain conditions while real-world locomotion is unstructured and more like bouts of intermittent steps. We use the energy minimization hypothesis to unify them under a single umbrella, and show gaits such as walk, trot and gallop in simple terrains, and unstructured gaits in rough terrains which is consistent with the findings in animal motor control. We validate our hypothesis in both simulation and real hardware across natural terrains.

*This was a HYBRID Event with in-person attendance in Levine 307 and Virtual attendance…