{"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/guiding-policies-with-language-via-meta","title":"Guiding Policies with Language via Meta-Learning","arxiv_id":"1811.07882","date":"2018-11-19","proceeding":"ICLR 2019 5","authors":["John D. Co-Reyes","Abhishek Gupta","Suvansh Sanjeev","Nick Altieri","Jacob Andreas","John DeNero","Pieter Abbeel","Sergey Levine"],"abstract":"Behavioral skills or policies for autonomous agents are conventionally\nlearned from reward functions, via reinforcement learning, or from\ndemonstrations, via imitation learning. However, both modes of task\nspecification have their disadvantages: reward functions require manual\nengineering, while demonstrations require a human expert to be able to actually\nperform the task in order to generate the demonstration. Instruction following\nfrom natural language instructions provides an appealing alternative: in the\nsame way that we can specify goals to other humans simply by speaking or\nwriting, we would like to be able to specify tasks for our machines. However, a\nsingle instruction may be insufficient to fully communicate our intent or, even\nif it is, may be insufficient for an autonomous agent to actually understand\nhow to perform the desired task. In this work, we propose an interactive\nformulation of the task specification problem, where iterative language\ncorrections are provided to an autonomous agent, guiding it in acquiring the\ndesired skill. Our proposed language-guided policy learning algorithm can\nintegrate an instruction and a sequence of corrections to acquire new skills\nvery quickly. In our experiments, we show that this method can enable a policy\nto follow instructions and corrections for simulated navigation and\nmanipulation tasks, substantially outperforming direct, non-interactive\ninstruction following.","url_abs":"http://arxiv.org/abs/1811.07882v2","url_pdf":"http://arxiv.org/pdf/1811.07882v2.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":"guiding-policies-with-language-via-meta","repo_url":"https://github.com/jcoreyes/lgpl","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"imitation-learning","task_name":"Imitation Learning"},{"task_slug":"instruction-following","task_name":"Instruction Following"},{"task_slug":"meta-learning","task_name":"Meta-Learning"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1811.07882","atlas_url":"https://app.syntology.ai/?focus=1811.07882","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}