Reinforcement Learning from Passive Data via Latent Intentions
This addresses the challenge of using untapped passive data for RL, potentially benefiting AI systems that rely on observational inputs, though it is incremental in applying intention modeling to this context.
The paper tackles the problem of leveraging abundant passive observational data, which lacks reward or action labels, for reinforcement learning by modeling intentions to learn features that accelerate downstream RL. The result is an algorithm that learns state, policy, and outcome representations from raw data, demonstrated to work with cross-embodiment video data and YouTube videos.
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.