{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/learning-symmetric-and-low-energy-locomotion","title":"Learning Symmetric and Low-energy Locomotion","arxiv_id":"1801.08093","date":"2018-01-24","proceeding":null,"authors":["Wenhao Yu","Greg Turk","C. Karen Liu"],"abstract":"Learning locomotion skills is a challenging problem. To generate realistic\nand smooth locomotion, existing methods use motion capture, finite state\nmachines or morphology-specific knowledge to guide the motion generation\nalgorithms. Deep reinforcement learning (DRL) is a promising approach for the\nautomatic creation of locomotion control. Indeed, a standard benchmark for DRL\nis to automatically create a running controller for a biped character from a\nsimple reward function. Although several different DRL algorithms can\nsuccessfully create a running controller, the resulting motions usually look\nnothing like a real runner. This paper takes a minimalist learning approach to\nthe locomotion problem, without the use of motion examples, finite state\nmachines, or morphology-specific knowledge. We introduce two modifications to\nthe DRL approach that, when used together, produce locomotion behaviors that\nare symmetric, low-energy, and much closer to that of a real person. First, we\nintroduce a new term to the loss function (not the reward function) that\nencourages symmetric actions. Second, we introduce a new curriculum learning\nmethod that provides modulated physical assistance to help the character with\nleft/right balance and forward movement. The algorithm automatically computes\nappropriate assistance to the character and gradually relaxes this assistance,\nso that eventually the character learns to move entirely without help. Because\nour method does not make use of motion capture data, it can be applied to a\nvariety of character morphologies. We demonstrate locomotion controllers for\nthe lower half of a biped, a full humanoid, a quadruped, and a hexapod. Our\nresults show that learned policies are able to produce symmetric, low-energy\ngaits. In addition, speed-appropriate gait patterns emerge without any guidance\nfrom motion examples or contact planning.","url_abs":"http://arxiv.org/abs/1801.08093v3","url_pdf":"http://arxiv.org/pdf/1801.08093v3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"learning-symmetric-and-low-energy-locomotion","repo_url":"https://github.com/VincentYu68/SymmetryCurriculumLocomotion","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null},{"paper_slug":"learning-symmetric-and-low-energy-locomotion","repo_url":"https://github.com/jyf588/lrle-rl-examples","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"deep-reinforcement-learning","task_name":"Deep Reinforcement Learning"},{"task_slug":"motion-generation","task_name":"Motion Generation"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1801.08093","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}