Learning to Walk with Rapid Motor Adaptation

Legged locomotion is commonly studied and programmed as a discrete set of structured gait patterns, like walk, trot, gallop.  However, studies of children learning to walk (Adolph et al) show that real-world locomotion is often quite unstructured and more like "bouts of intermittent steps". We have developed a general approach to walking which is built on learning on varied terrains in simulation and then fast online adaptation (fractions of a second) in the real world. This is made possible by our  Rapid Motor Adaptation (RMA) algorithm. RMA consists of two components: a base policy and an adaptation module, both of which can be trained in simulation.  We thus learn walking policies that are much more flexible and adaptable. In our set-up gaits emerge as a consequence of minimizing energy consumption at different target speeds, consistent with various animal motor studies. You can see our robot walking at https://www.youtube.com/watch?v=nBy1piJrq1A