Score and Distribution Matching Policy: Advanced Accelerated Visuomotor Policies via Matched Distillation

TL;DR

This paper introduces the SDM Policy, a novel method that transforms diffusion-based visual-motor policies into fast, single-step generators, achieving high-quality actions with significantly reduced inference time for robotic control.

Contribution

The paper presents the SDM Policy, combining score and distribution matching with a dual-teacher mechanism to improve inference speed and action quality in diffusion-based policies.

Findings

6x faster inference speed on a 57-task benchmark

State-of-the-art action quality achieved

Robust and reliable high-frequency control performance

Abstract

Visual-motor policy learning has advanced with architectures like diffusion-based policies, known for modeling complex robotic trajectories. However, their prolonged inference times hinder high-frequency control tasks requiring real-time feedback. While consistency distillation (CD) accelerates inference, it introduces errors that compromise action quality. To address these limitations, we propose the Score and Distribution Matching Policy (SDM Policy), which transforms diffusion-based policies into single-step generators through a two-stage optimization process: score matching ensures alignment with true action distributions, and distribution matching minimizes KL divergence for consistency. A dual-teacher mechanism integrates a frozen teacher for stability and an unfrozen teacher for adversarial training, enhancing robustness and alignment with target distributions. Evaluated on a 57-task simulation benchmark, SDM Policy achieves a 6x inference speedup while having state-of-the-art action quality, providing an efficient and reliable framework for high-frequency robotic tasks.