Self-Guided Action Diffusion

TL;DR

This paper introduces self-guided action diffusion, a more efficient inference-time search method for diffusion-based robot policies that significantly improves success rates under limited sampling budgets.

Contribution

It proposes a novel self-guided approach that reduces computational costs while maintaining high performance in diffusion-based policy inference.

Findings

Achieves up to 70% higher success rates under tight sampling budgets.

Enables near-optimal performance with negligible inference cost.

Improves the efficiency of bidirectional decoding in diffusion policies.

Abstract

Recent works have shown the promise of inference-time search over action samples for improving generative robot policies. In particular, optimizing cross-chunk coherence via bidirectional decoding has proven effective in boosting the consistency and reactivity of diffusion policies. However, this approach remains computationally expensive as the diversity of sampled actions grows. In this paper, we introduce self-guided action diffusion, a more efficient variant of bidirectional decoding tailored for diffusion-based policies. At the core of our method is to guide the proposal distribution at each diffusion step based on the prior decision. Experiments in simulation tasks show that the proposed self-guidance enables near-optimal performance at negligible inference cost. Notably, under a tight sampling budget, our method achieves up to 70% higher success rates than existing counterparts on challenging dynamic tasks. See project website at https://rhea-mal.github.io/selfgad.github.io.