{"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/universal-planning-networks","title":"Universal Planning Networks","arxiv_id":"1804.00645","date":"2018-04-02","proceeding":null,"authors":["Aravind Srinivas","Allan Jabri","Pieter Abbeel","Sergey Levine","Chelsea Finn"],"abstract":"A key challenge in complex visuomotor control is learning abstract\nrepresentations that are effective for specifying goals, planning, and\ngeneralization. To this end, we introduce universal planning networks (UPN).\nUPNs embed differentiable planning within a goal-directed policy. This planning\ncomputation unrolls a forward model in a latent space and infers an optimal\naction plan through gradient descent trajectory optimization. The\nplan-by-gradient-descent process and its underlying representations are learned\nend-to-end to directly optimize a supervised imitation learning objective. We\nfind that the representations learned are not only effective for goal-directed\nvisual imitation via gradient-based trajectory optimization, but can also\nprovide a metric for specifying goals using images. The learned representations\ncan be leveraged to specify distance-based rewards to reach new target states\nfor model-free reinforcement learning, resulting in substantially more\neffective learning when solving new tasks described via image-based goals. We\nwere able to achieve successful transfer of visuomotor planning strategies\nacross robots with significantly different morphologies and actuation\ncapabilities.","url_abs":"http://arxiv.org/abs/1804.00645v2","url_pdf":"http://arxiv.org/pdf/1804.00645v2.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":"universal-planning-networks","repo_url":"https://github.com/aravindsrinivas/upn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"imitation-learning","task_name":"Imitation Learning"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"representation-learning","task_name":"Representation Learning"},{"task_slug":"transfer-reinforcement-learning","task_name":"Transfer Reinforcement Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1804.00645","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}