Legged locomotion is commonly studied and expressed as a discrete set of gait\npatterns, like walk, trot, gallop, which are usually treated as given and\npre-programmed in legged robots for efficient locomotion at different speeds.\nHowever, fixing a set of pre-programmed gaits limits the generality of\nlocomotion. Recent animal motor studies show that these conventional gaits are\nonly prevalent in ideal flat terrain conditions while real-world locomotion is\nunstructured and more like bouts of intermittent steps. What principles could\nlead to both structured and unstructured patterns across mammals and how to\nsynthesize them in robots? In this work, we take an analysis-by-synthesis\napproach and learn to move by minimizing mechanical energy. We demonstrate that\nlearning to minimize energy consumption plays a key role in the emergence of\nnatural locomotion gaits at different speeds in real quadruped robots. The\nemergent gaits are structured in ideal terrains and look similar to that of\nhorses and sheep. The same approach leads to unstructured gaits in rough\nterrains which is consistent with the findings in animal motor control. We\nvalidate our hypothesis in both simulation and real hardware across natural\nterrains. Videos at https://energy-locomotion.github.io\n
Discussion(0)
No comments yet. Be the first to comment.