{"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/rail-risk-averse-imitation-learning","title":"RAIL: Risk-Averse Imitation Learning","arxiv_id":"1707.06658","date":"2017-07-20","proceeding":null,"authors":["Anirban Santara","Abhishek Naik","Balaraman Ravindran","Dipankar Das","Dheevatsa Mudigere","Sasikanth Avancha","Bharat Kaul"],"abstract":"Imitation learning algorithms learn viable policies by imitating an expert's\nbehavior when reward signals are not available. Generative Adversarial\nImitation Learning (GAIL) is a state-of-the-art algorithm for learning policies\nwhen the expert's behavior is available as a fixed set of trajectories. We\nevaluate in terms of the expert's cost function and observe that the\ndistribution of trajectory-costs is often more heavy-tailed for GAIL-agents\nthan the expert at a number of benchmark continuous-control tasks. Thus,\nhigh-cost trajectories, corresponding to tail-end events of catastrophic\nfailure, are more likely to be encountered by the GAIL-agents than the expert.\nThis makes the reliability of GAIL-agents questionable when it comes to\ndeployment in risk-sensitive applications like robotic surgery and autonomous\ndriving. In this work, we aim to minimize the occurrence of tail-end events by\nminimizing tail risk within the GAIL framework. We quantify tail risk by the\nConditional-Value-at-Risk (CVaR) of trajectories and develop the Risk-Averse\nImitation Learning (RAIL) algorithm. We observe that the policies learned with\nRAIL show lower tail-end risk than those of vanilla GAIL. Thus the proposed\nRAIL algorithm appears as a potent alternative to GAIL for improved reliability\nin risk-sensitive applications.","url_abs":"http://arxiv.org/abs/1707.06658v4","url_pdf":"http://arxiv.org/pdf/1707.06658v4.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":"rail-risk-averse-imitation-learning","repo_url":"https://github.com/Santara/RAIL","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"autonomous-driving","task_name":"Autonomous Driving"},{"task_slug":"continuous-control","task_name":"Continuous Control"},{"task_slug":"imitation-learning","task_name":"Imitation Learning"},{"task_slug":"continuous-control","task_name":"continuous-control"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1707.06658","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}