{"day":"31","date_created":"2023-08-22T14:23:16Z","type":"preprint","year":"2022","publication":"arXiv","language":[{"iso":"eng"}],"doi":"10.48550/arXiv.2201.13388","status":"public","citation":{"mla":"Mambelli, Davide, et al. “Compositional Multi-Object Reinforcement Learning with Linear Relation Networks.” ArXiv, 2201.13388, doi:10.48550/arXiv.2201.13388.","chicago":"Mambelli, Davide, Frederik Träuble, Stefan Bauer, Bernhard Schölkopf, and Francesco Locatello. “Compositional Multi-Object Reinforcement Learning with Linear Relation Networks.” ArXiv, n.d. https://doi.org/10.48550/arXiv.2201.13388.","short":"D. Mambelli, F. Träuble, S. Bauer, B. Schölkopf, F. Locatello, ArXiv (n.d.).","ieee":"D. Mambelli, F. Träuble, S. Bauer, B. Schölkopf, and F. Locatello, “Compositional multi-object reinforcement learning with linear relation networks,” arXiv. .","ama":"Mambelli D, Träuble F, Bauer S, Schölkopf B, Locatello F. Compositional multi-object reinforcement learning with linear relation networks. arXiv. doi:10.48550/arXiv.2201.13388","ista":"Mambelli D, Träuble F, Bauer S, Schölkopf B, Locatello F. Compositional multi-object reinforcement learning with linear relation networks. arXiv, 2201.13388.","apa":"Mambelli, D., Träuble, F., Bauer, S., Schölkopf, B., & Locatello, F. (n.d.). Compositional multi-object reinforcement learning with linear relation networks. arXiv. https://doi.org/10.48550/arXiv.2201.13388"},"department":[{"_id":"FrLo"}],"main_file_link":[{"url":"https://doi.org/10.48550/arXiv.2201.13388","open_access":"1"}],"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","oa":1,"article_processing_charge":"No","date_published":"2022-01-31T00:00:00Z","_id":"14220","external_id":{"arxiv":["2201.13388"]},"date_updated":"2024-10-14T12:27:39Z","publication_status":"submitted","article_number":"2201.13388","abstract":[{"text":"Although reinforcement learning has seen remarkable progress over the last years, solving robust dexterous object-manipulation tasks in multi-object settings remains a challenge. In this paper, we focus on models that can learn manipulation tasks in fixed multi-object settings and extrapolate this skill zero-shot without any drop in performance when the number of objects changes. We consider the generic task of bringing a specific cube out of a set to a goal position. We find that previous approaches, which primarily leverage attention and graph neural network-based architectures, do not generalize their skills when the number of input objects changes while scaling as K2. We propose an alternative plug-and-play module based on relational inductive biases to overcome these limitations. Besides exceeding performances in their training environment, we show that our approach, which scales linearly in K, allows agents to extrapolate and generalize zero-shot to any new object number.","lang":"eng"}],"title":"Compositional multi-object reinforcement learning with linear relation networks","extern":"1","author":[{"first_name":"Davide","full_name":"Mambelli, Davide","last_name":"Mambelli"},{"first_name":"Frederik","full_name":"Träuble, Frederik","last_name":"Träuble"},{"last_name":"Bauer","full_name":"Bauer, Stefan","first_name":"Stefan"},{"full_name":"Schölkopf, Bernhard","last_name":"Schölkopf","first_name":"Bernhard"},{"last_name":"Locatello","full_name":"Locatello, Francesco","orcid":"0000-0002-4850-0683","first_name":"Francesco","id":"26cfd52f-2483-11ee-8040-88983bcc06d4"}],"month":"01","oa_version":"Preprint"}