Responsive Noise-Relaying Diffusion Policy: Responsive and Efficient Visuomotor Control

2025-02-18

AI Summary

Existing diffusion-based policies for imitation learning generate multi-step actions, resulting in high latency and failing to meet the real-time requirements of robotic visuomotor control. To address this, we propose a noise-relay buffering mechanism coupled with sequential denoising diffusion. Our approach introduces a novel noise-relay buffer that jointly enables instantaneous denoising at the head and progressive denoising at the tail, preserving action continuity while enhancing single-step responsiveness to the latest visual observations. We further incorporate a conditional denoising diffusion model, serialized progressive noise scheduling, and noise-buffer reuse to achieve end-to-end vision-to-action mapping. Experiments demonstrate an 18% improvement in success rate on latency-sensitive tasks and a 6.9% increase in inference speed over the best prior acceleration method on standard tasks—effectively balancing real-time performance and motion consistency.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Imitation learning is an efficient method for teaching robots a variety of tasks. Diffusion Policy, which uses a conditional denoising diffusion process to generate actions, has demonstrated superior performance, particularly in learning from multi-modal demonstrates. However, it relies on executing multiple actions to retain performance and prevent mode bouncing, which limits its responsiveness, as actions are not conditioned on the most recent observations. To address this, we introduce Responsive Noise-Relaying Diffusion Policy (RNR-DP), which maintains a noise-relaying buffer with progressively increasing noise levels and employs a sequential denoising mechanism that generates immediate, noise-free actions at the head of the sequence, while appending noisy actions at the tail. This ensures that actions are responsive and conditioned on the latest observations, while maintaining motion consistency through the noise-relaying buffer. This design enables the handling of tasks requiring responsive control, and accelerates action generation by reusing denoising steps. Experiments on response-sensitive tasks demonstrate that, compared to Diffusion Policy, ours achieves 18% improvement in success rate. Further evaluation on regular tasks demonstrates that RNR-DP also exceeds the best acceleration method by 6.9%, highlighting its computational efficiency advantage in scenarios where responsiveness is less critical.

Problem

• Improving robot action responsiveness
• Enhancing visuomotor control efficiency
• Reducing computational delay in tasks

Innovation

• Utilizes noise-relaying buffer
• Sequential denoising mechanism
• Improves action responsiveness