Scaling Offline RL via Efficient and Expressive Shortcut Models
This addresses the computational bottleneck of iterative noise sampling in generative models for offline RL, offering a more efficient approach for practitioners.
The paper tackles the challenge of applying diffusion and flow models to offline reinforcement learning by introducing SORL, a new algorithm that uses shortcut models to scale training and inference. SORL achieves strong performance across offline RL tasks and shows positive scaling with increased test-time compute.
Diffusion and flow models have emerged as powerful generative approaches capable of modeling diverse and multimodal behavior. However, applying these models to offline reinforcement learning (RL) remains challenging due to the iterative nature of their noise sampling processes, making policy optimization difficult. In this paper, we introduce Scalable Offline Reinforcement Learning (SORL), a new offline RL algorithm that leverages shortcut models - a novel class of generative models - to scale both training and inference. SORL's policy can capture complex data distributions and can be trained simply and efficiently in a one-stage training procedure. At test time, SORL introduces both sequential and parallel inference scaling by using the learned Q-function as a verifier. We demonstrate that SORL achieves strong performance across a range of offline RL tasks and exhibits positive scaling behavior with increased test-time compute. We release the code at nico-espinosadice.github.io/projects/sorl.