Diffusion Reward: Learning Rewards via Conditional Video Diffusion
This work addresses the challenge of specifying behaviors in visual reinforcement learning for robotics, offering an incremental improvement by applying diffusion models to reward learning from videos.
The authors tackled the problem of learning rewards from expert videos for reinforcement learning tasks by proposing Diffusion Reward, a framework that uses conditional video diffusion models to encourage exploration of expert behaviors, achieving success in robotic manipulation tasks in simulation and real-world settings and outperforming baseline methods on unseen tasks.
Learning rewards from expert videos offers an affordable and effective solution to specify the intended behaviors for reinforcement learning (RL) tasks. In this work, we propose Diffusion Reward, a novel framework that learns rewards from expert videos via conditional video diffusion models for solving complex visual RL problems. Our key insight is that lower generative diversity is exhibited when conditioning diffusion on expert trajectories. Diffusion Reward is accordingly formalized by the negative of conditional entropy that encourages productive exploration of expert behaviors. We show the efficacy of our method over robotic manipulation tasks in both simulation platforms and the real world with visual input. Moreover, Diffusion Reward can even solve unseen tasks successfully and effectively, largely surpassing baseline methods. Project page and code: https://diffusion-reward.github.io.