{"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/hierarchical-reinforcement-learning-via","title":"Hierarchical Reinforcement Learning via Advantage-Weighted Information Maximization","arxiv_id":"1901.01365","date":"2019-01-05","proceeding":"ICLR 2019 5","authors":["Takayuki Osa","Voot Tangkaratt","Masashi Sugiyama"],"abstract":"Real-world tasks are often highly structured. Hierarchical reinforcement\nlearning (HRL) has attracted research interest as an approach for leveraging\nthe hierarchical structure of a given task in reinforcement learning (RL).\nHowever, identifying the hierarchical policy structure that enhances the\nperformance of RL is not a trivial task. In this paper, we propose an HRL\nmethod that learns a latent variable of a hierarchical policy using mutual\ninformation maximization. Our approach can be interpreted as a way to learn a\ndiscrete and latent representation of the state-action space. To learn option\npolicies that correspond to modes of the advantage function, we introduce\nadvantage-weighted importance sampling. In our HRL method, the gating policy\nlearns to select option policies based on an option-value function, and these\noption policies are optimized based on the deterministic policy gradient\nmethod. This framework is derived by leveraging the analogy between a\nmonolithic policy in standard RL and a hierarchical policy in HRL by using a\ndeterministic option policy. Experimental results indicate that our HRL\napproach can learn a diversity of options and that it can enhance the\nperformance of RL in continuous control tasks.","url_abs":"http://arxiv.org/abs/1901.01365v2","url_pdf":"http://arxiv.org/pdf/1901.01365v2.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":"hierarchical-reinforcement-learning-via","repo_url":"https://github.com/TakaOsa/adInfoHRL","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"continuous-control","task_name":"Continuous Control"},{"task_slug":"diversity","task_name":"Diversity"},{"task_slug":"hierarchical-reinforcement-learning","task_name":"Hierarchical 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"},{"task_slug":"reinforcement-learning-2","task_name":"reinforcement-learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}