{"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/archer-aggressive-rewards-to-counter-bias-in","title":"ARCHER: Aggressive Rewards to Counter bias in Hindsight Experience Replay","arxiv_id":"1809.02070","date":"2018-09-06","proceeding":null,"authors":["Sameera Lanka","Tianfu Wu"],"abstract":"Experience replay is an important technique for addressing\nsample-inefficiency in deep reinforcement learning (RL), but faces difficulty\nin learning from binary and sparse rewards due to disproportionately few\nsuccessful experiences in the replay buffer. Hindsight experience replay (HER)\nwas recently proposed to tackle this difficulty by manipulating unsuccessful\ntransitions, but in doing so, HER introduces a significant bias in the replay\nbuffer experiences and therefore achieves a suboptimal improvement in\nsample-efficiency. In this paper, we present an analysis on the source of bias\nin HER, and propose a simple and effective method to counter the bias, to most\neffectively harness the sample-efficiency provided by HER. Our method,\nmotivated by counter-factual reasoning and called ARCHER, extends HER with a\ntrade-off to make rewards calculated for hindsight experiences numerically\ngreater than real rewards. We validate our algorithm on two continuous control\nenvironments from DeepMind Control Suite - Reacher and Finger, which simulate\nmanipulation tasks with a robotic arm - in combination with various reward\nfunctions, task complexities and goal sampling strategies. Our experiments\nconsistently demonstrate that countering bias using more aggressive hindsight\nrewards increases sample efficiency, thus establishing the greater benefit of\nARCHER in RL applications with limited computing budget.","url_abs":"http://arxiv.org/abs/1809.02070v2","url_pdf":"http://arxiv.org/pdf/1809.02070v2.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":"archer-aggressive-rewards-to-counter-bias-in","repo_url":"https://github.com/Baichenjia/BHER","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"continuous-control","task_name":"Continuous Control"},{"task_slug":"deep-reinforcement-learning","task_name":"Deep Reinforcement Learning"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"},{"task_slug":"continuous-control","task_name":"continuous-control"}],"methods":[{"method_slug":"experience-replay","method_name":"Experience Replay"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1809.02070","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}