Improving Diffusion Planners by Self-Supervised Action Gating with Energies

TL;DR

This paper introduces SAGE, a self-supervised re-ranking method that enhances diffusion planners in offline reinforcement learning by penalizing dynamically inconsistent plans through a latent energy score, improving robustness and performance.

Contribution

SAGE is a novel inference-time re-ranking approach that uses a joint-embedding predictive architecture to evaluate and penalize inconsistent trajectories without additional environment interactions.

Findings

SAGE improves performance across locomotion, navigation, and manipulation benchmarks.

SAGE enhances robustness of diffusion planners without environment rollouts.

SAGE integrates seamlessly into existing diffusion planning pipelines.

Abstract

Diffusion planners are a strong approach for offline reinforcement learning, but they can fail when value-guided selection favours trajectories that score well yet are locally inconsistent with the environment dynamics, resulting in brittle execution. We propose Self-supervised Action Gating with Energies (SAGE), an inference-time re-ranking method that penalises dynamically inconsistent plans using a latent consistency signal. SAGE trains a Joint-Embedding Predictive Architecture (JEPA) encoder on offline state sequences and an action-conditioned latent predictor for short horizon transitions. At test time, SAGE assigns each sampled candidate an energy given by its latent prediction error and combines this feasibility score with value estimates to select actions. SAGE can integrate into existing diffusion planning pipelines that can sample trajectories and select actions via value scoring; it requires no environment rollouts and no policy re-training. Across locomotion, navigation, and manipulation benchmarks, SAGE improves the performance and robustness of diffusion planners.