World Models via Policy-Guided Trajectory Diffusion
This addresses the bottleneck of error accumulation in world models for RL agents, offering a more efficient and accurate alternative for synthetic data generation.
The paper tackles the problem of compounding prediction errors in autoregressive world models for reinforcement learning by proposing PolyGRAD, a non-autoregressive method that generates entire on-policy trajectories using diffusion. The results show that PolyGRAD outperforms state-of-the-art baselines in trajectory prediction error for short trajectories, with similar errors to autoregressive diffusion but lower computational costs, and enables performant policy training in MuJoCo domains.
World models are a powerful tool for developing intelligent agents. By predicting the outcome of a sequence of actions, world models enable policies to be optimised via on-policy reinforcement learning (RL) using synthetic data, i.e. in "in imagination". Existing world models are autoregressive in that they interleave predicting the next state with sampling the next action from the policy. Prediction error inevitably compounds as the trajectory length grows. In this work, we propose a novel world modelling approach that is not autoregressive and generates entire on-policy trajectories in a single pass through a diffusion model. Our approach, Policy-Guided Trajectory Diffusion (PolyGRAD), leverages a denoising model in addition to the gradient of the action distribution of the policy to diffuse a trajectory of initially random states and actions into an on-policy synthetic trajectory. We analyse the connections between PolyGRAD, score-based generative models, and classifier-guided diffusion models. Our results demonstrate that PolyGRAD outperforms state-of-the-art baselines in terms of trajectory prediction error for short trajectories, with the exception of autoregressive diffusion. For short trajectories, PolyGRAD obtains similar errors to autoregressive diffusion, but with lower computational requirements. For long trajectories, PolyGRAD obtains comparable performance to baselines. Our experiments demonstrate that PolyGRAD enables performant policies to be trained via on-policy RL in imagination for MuJoCo continuous control domains. Thus, PolyGRAD introduces a new paradigm for accurate on-policy world modelling without autoregressive sampling.