DC Field | Value | Language |
---|---|---|
dc.contributor.author | Seo, Younggyo | ko |
dc.contributor.author | Lee, Kimin | ko |
dc.contributor.author | James, Stephen | ko |
dc.contributor.author | Abbeel, Pieter | ko |
dc.date.accessioned | 2023-05-10T10:00:35Z | - |
dc.date.available | 2023-05-10T10:00:35Z | - |
dc.date.created | 2023-05-03 | - |
dc.date.created | 2023-05-03 | - |
dc.date.issued | 2022-07-20 | - |
dc.identifier.citation | 38th International Conference on Machine Learning (ICML), pp.19561 - 19579 | - |
dc.identifier.issn | 2640-3498 | - |
dc.identifier.uri | http://hdl.handle.net/10203/306680 | - |
dc.description.abstract | Recent unsupervised pre-training methods have shown to be effective on language and vision domains by learning useful representations for multiple downstream tasks. In this paper, we investigate if such unsupervised pre-training methods can also be effective for vision-based reinforcement learning (RL). To this end, we introduce a framework that learns representations useful for understanding the dynamics via generative pre-training on videos. Our framework consists of two phases: we pre-train an action-free latent video prediction model, and then utilize the pre-trained representations for efficiently learning action-conditional world models on unseen environments. To incorporate additional action inputs during fine-tuning, we introduce a new architecture that stacks an action-conditional latent prediction model on top of the pre-trained action-free prediction model. Moreover, for better exploration, we propose a video-based intrinsic bonus that leverages pre-trained representations. We demonstrate that our framework significantly improves both final performances and sample-efficiency of vision-based RL in a variety of manipulation and locomotion tasks. Code is available at https://github.com/younggyoseo/apv. | - |
dc.language | English | - |
dc.publisher | JMLR-JOURNAL MACHINE LEARNING RESEARCH | - |
dc.title | Reinforcement Learning with Action-Free Pre-Training from Videos | - |
dc.type | Conference | - |
dc.identifier.wosid | 000900130200026 | - |
dc.type.rims | CONF | - |
dc.citation.beginningpage | 19561 | - |
dc.citation.endingpage | 19579 | - |
dc.citation.publicationname | 38th International Conference on Machine Learning (ICML) | - |
dc.identifier.conferencecountry | US | - |
dc.identifier.conferencelocation | Baltimore, MD | - |
dc.contributor.localauthor | Lee, Kimin | - |
dc.contributor.nonIdAuthor | James, Stephen | - |
dc.contributor.nonIdAuthor | Abbeel, Pieter | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.