Towards fast learning of robotic locomotion

Robots are a key part of our future; they will need to be able to move and work in various environments and with changing interactions with objects and environment. A big part of today’s research focuses on different aspects of making robots part of our lives. One aspect is the ability of a robot to move in various terrains, on slopes, slippery terrain, and even dynamically changing terrain like sand and gravel. Our research shows a method for robots to rapidly learn their new environments using geometric mechanics concepts. A geometric mechanics analysis of robotic locomotion has provided valuable insights into how biological and robotic systems use changes in their internal shape in order to generate net movement by exploiting mechanical interaction with the environment. This perspective produced an approach for obtaining simplified data-driven models of locomotion systems directly from motion tracking data.