{"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/object-oriented-dynamics-predictor","title":"Object-Oriented Dynamics Predictor","arxiv_id":"1806.07371","date":"2018-05-25","proceeding":"NeurIPS 2018 12","authors":["Guangxiang Zhu","Zhiao Huang","Chongjie Zhang"],"abstract":"Generalization has been one of the major challenges for learning dynamics\nmodels in model-based reinforcement learning. However, previous work on\naction-conditioned dynamics prediction focuses on learning the pixel-level\nmotion and thus does not generalize well to novel environments with different\nobject layouts. In this paper, we present a novel object-oriented framework,\ncalled object-oriented dynamics predictor (OODP), which decomposes the\nenvironment into objects and predicts the dynamics of objects conditioned on\nboth actions and object-to-object relations. It is an end-to-end neural network\nand can be trained in an unsupervised manner. To enable the generalization\nability of dynamics learning, we design a novel CNN-based relation mechanism\nthat is class-specific (rather than object-specific) and exploits the locality\nprinciple. Empirical results show that OODP significantly outperforms previous\nmethods in terms of generalization over novel environments with various object\nlayouts. OODP is able to learn from very few environments and accurately\npredict dynamics in a large number of unseen environments. In addition, OODP\nlearns semantically and visually interpretable dynamics models.","url_abs":"http://arxiv.org/abs/1806.07371v3","url_pdf":"http://arxiv.org/pdf/1806.07371v3.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":"object-oriented-dynamics-predictor","repo_url":"https://github.com/mig-zh/OODP","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"model-based-reinforcement-learning","task_name":"Model-based Reinforcement Learning"},{"task_slug":"object","task_name":"Object"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1806.07371","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}