{"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/a-compositional-object-based-approach-to","title":"A Compositional Object-Based Approach to Learning Physical Dynamics","arxiv_id":"1612.00341","date":"2016-12-01","proceeding":null,"authors":["Michael B. Chang","Tomer Ullman","Antonio Torralba","Joshua B. Tenenbaum"],"abstract":"We present the Neural Physics Engine (NPE), a framework for learning\nsimulators of intuitive physics that naturally generalize across variable\nobject count and different scene configurations. We propose a factorization of\na physical scene into composable object-based representations and a neural\nnetwork architecture whose compositional structure factorizes object dynamics\ninto pairwise interactions. Like a symbolic physics engine, the NPE is endowed\nwith generic notions of objects and their interactions; realized as a neural\nnetwork, it can be trained via stochastic gradient descent to adapt to specific\nobject properties and dynamics of different worlds. We evaluate the efficacy of\nour approach on simple rigid body dynamics in two-dimensional worlds. By\ncomparing to less structured architectures, we show that the NPE's\ncompositional representation of the structure in physical interactions improves\nits ability to predict movement, generalize across variable object count and\ndifferent scene configurations, and infer latent properties of objects such as\nmass.","url_abs":"http://arxiv.org/abs/1612.00341v2","url_pdf":"http://arxiv.org/pdf/1612.00341v2.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":"a-compositional-object-based-approach-to","repo_url":"https://github.com/mbchang/dynamics","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"torch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"object","task_name":"Object"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1612.00341","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}