Reinforcement Learning from Passive Data via Latent Intentions

Dibya Ghosh, Chethan Anand Bhateja, Sergey Levine
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:11321-11339, 2023.

Abstract

Passive observational data, such as human videos, is abundant and rich in information, yet remains largely untapped by current RL methods. Perhaps surprisingly, we show that passive data, despite not having reward or action labels, can still be used to learn features that accelerate downstream RL. Our approach learns from passive data by modeling intentions: measuring how the likelihood of future outcomes change when the agent acts to achieve a particular task. We propose a temporal difference learning objective to learn about intentions, resulting in an algorithm similar to conventional RL, but which learns entirely from passive data. When optimizing this objective, our agent simultaneously learns representations of states, of policies, and of possible outcomes in an environment, all from raw observational data. Both theoretically and empirically, this scheme learns features amenable for value prediction for downstream tasks, and our experiments demonstrate the ability to learn from many forms of passive data, including cross-embodiment video data and YouTube videos.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-ghosh23a, title = {Reinforcement Learning from Passive Data via Latent Intentions}, author = {Ghosh, Dibya and Bhateja, Chethan Anand and Levine, Sergey}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {11321--11339}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/ghosh23a/ghosh23a.pdf}, url = {https://proceedings.mlr.press/v202/ghosh23a.html}, abstract = {Passive observational data, such as human videos, is abundant and rich in information, yet remains largely untapped by current RL methods. Perhaps surprisingly, we show that passive data, despite not having reward or action labels, can still be used to learn features that accelerate downstream RL. Our approach learns from passive data by modeling intentions: measuring how the likelihood of future outcomes change when the agent acts to achieve a particular task. We propose a temporal difference learning objective to learn about intentions, resulting in an algorithm similar to conventional RL, but which learns entirely from passive data. When optimizing this objective, our agent simultaneously learns representations of states, of policies, and of possible outcomes in an environment, all from raw observational data. Both theoretically and empirically, this scheme learns features amenable for value prediction for downstream tasks, and our experiments demonstrate the ability to learn from many forms of passive data, including cross-embodiment video data and YouTube videos.} }
Endnote
%0 Conference Paper %T Reinforcement Learning from Passive Data via Latent Intentions %A Dibya Ghosh %A Chethan Anand Bhateja %A Sergey Levine %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-ghosh23a %I PMLR %P 11321--11339 %U https://proceedings.mlr.press/v202/ghosh23a.html %V 202 %X Passive observational data, such as human videos, is abundant and rich in information, yet remains largely untapped by current RL methods. Perhaps surprisingly, we show that passive data, despite not having reward or action labels, can still be used to learn features that accelerate downstream RL. Our approach learns from passive data by modeling intentions: measuring how the likelihood of future outcomes change when the agent acts to achieve a particular task. We propose a temporal difference learning objective to learn about intentions, resulting in an algorithm similar to conventional RL, but which learns entirely from passive data. When optimizing this objective, our agent simultaneously learns representations of states, of policies, and of possible outcomes in an environment, all from raw observational data. Both theoretically and empirically, this scheme learns features amenable for value prediction for downstream tasks, and our experiments demonstrate the ability to learn from many forms of passive data, including cross-embodiment video data and YouTube videos.
APA
Ghosh, D., Bhateja, C.A. & Levine, S.. (2023). Reinforcement Learning from Passive Data via Latent Intentions. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:11321-11339 Available from https://proceedings.mlr.press/v202/ghosh23a.html.

Related Material