Planning with Diffusion for Flexible Behavior Synthesis
This work addresses the challenge of improving planning flexibility in reinforcement learning for robotics or control applications, though it appears incremental by adapting diffusion models to a known bottleneck.
The paper tackles the problem of model-based reinforcement learning by proposing a diffusion probabilistic model that integrates trajectory optimization into the modeling process, enabling flexible behavior synthesis through iterative denoising of trajectories, and demonstrates effectiveness in control settings requiring long-horizon decision-making.
Model-based reinforcement learning methods often use learning only for the purpose of estimating an approximate dynamics model, offloading the rest of the decision-making work to classical trajectory optimizers. While conceptually simple, this combination has a number of empirical shortcomings, suggesting that learned models may not be well-suited to standard trajectory optimization. In this paper, we consider what it would look like to fold as much of the trajectory optimization pipeline as possible into the modeling problem, such that sampling from the model and planning with it become nearly identical. The core of our technical approach lies in a diffusion probabilistic model that plans by iteratively denoising trajectories. We show how classifier-guided sampling and image inpainting can be reinterpreted as coherent planning strategies, explore the unusual and useful properties of diffusion-based planning methods, and demonstrate the effectiveness of our framework in control settings that emphasize long-horizon decision-making and test-time flexibility.