DynamicManip: Enabling Dynamic Manipulation from a Single Static Demonstration

2026-08-02

AI Summary

This work addresses the challenges of high data requirements and policy inference latency that often hinder robotic manipulation in dynamic environments. The authors propose a novel augmentation method capable of generating diverse dynamic training data from a single static demonstration, coupled with a dynamics-aware adaptive policy that modulates its inference frequency based on environmental changes to significantly reduce latency. The study introduces the first benchmark for dynamic manipulation tasks along with an automated evaluation framework, enabling efficient and low-latency imitation learning. Experimental results demonstrate that the proposed approach improves average task success rates by 18.4% and reduces policy query latency by 32.9% in both simulation and real-world settings.

Abstract

Dynamic manipulation is a critical capability for robots operating in complex and dynamic environments, where robots must interact with objects that are moving or require rapid adjustments. However, learning models for dynamic manipulation tasks face two major challenges: (1) the combinatorial complexity of dynamic scenarios leads to substantial data requirements, and (2) rapid variations in dynamics require real-time and accurate policy execution. In this paper, we propose DynamicManip to address these challenges through an efficient data augmentation pipeline and a low-latency imitation policy. We first propose a static-to-dynamic augmentation pipeline that synthesizes diverse dynamic manipulation demonstrations from a single static demonstration. Second, we introduce a dynamic-aware adaptive policy that adjusts its inference frequency according to task dynamics, enabling responsive and effective dynamic manipulation. Third, we build a dynamic manipulation benchmark, which includes diverse dynamic tasks with an automatic evaluation system for scalable and consistent assessment. Extensive experiments in both simulation and the real world demonstrate that DynamicManip not only provides significant improvements in data efficiency but also achieves better performance in dynamic manipulation tasks, with a mean success rate 18.4 percentage points higher and policy-query latency 32.9% lower.

Problem

• dynamic manipulation
• data efficiency
• real-time policy
• robot learning
• imitation learning

Innovation

• dynamic manipulation
• data augmentation
• adaptive policy
• imitation learning
• robotics benchmark