Deep Reactive Policy: Learning Reactive Manipulator Motion Planning for Dynamic Environments

2025-09-08

AI Summary

To address the challenge of real-time, collision-free robotic arm motion planning in dynamic and partially observable environments, this paper proposes a vision–motor neural policy that directly processes raw point clouds to generate reactive, low-latency trajectories. Methodologically, we pretrain the IMPACT foundation model on millions of expert demonstrations from simulation, integrating a point-cloud encoder, Transformer architecture, and iterative teacher–student fine-tuning. We further introduce a Dynamic Control Point–Reciprocal Motion Planner (DCP-RMP) module to enable online target re-planning during inference. Compared to conventional planners—reliant on global environment modeling and suffering from high computational latency—and existing neural policies—limited in generalization—our approach achieves significant improvements in success rate, robustness, and cross-environment generalization, both in simulation and on physical hardware. To our knowledge, this is the first end-to-end dynamic obstacle avoidance framework delivering high-precision control with sub-100ms latency and strong adaptability to unseen dynamics.

Technology Category

• Intelligent Robots: Motion and Path Planning
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Generating collision-free motion in dynamic, partially observable environments is a fundamental challenge for robotic manipulators. Classical motion planners can compute globally optimal trajectories but require full environment knowledge and are typically too slow for dynamic scenes. Neural motion policies offer a promising alternative by operating in closed-loop directly on raw sensory inputs but often struggle to generalize in complex or dynamic settings. We propose Deep Reactive Policy (DRP), a visuo-motor neural motion policy designed for reactive motion generation in diverse dynamic environments, operating directly on point cloud sensory input. At its core is IMPACT, a transformer-based neural motion policy pretrained on 10 million generated expert trajectories across diverse simulation scenarios. We further improve IMPACT's static obstacle avoidance through iterative student-teacher finetuning. We additionally enhance the policy's dynamic obstacle avoidance at inference time using DCP-RMP, a locally reactive goal-proposal module. We evaluate DRP on challenging tasks featuring cluttered scenes, dynamic moving obstacles, and goal obstructions. DRP achieves strong generalization, outperforming prior classical and neural methods in success rate across both simulated and real-world settings. Video results and code available at https://deep-reactive-policy.com

Problem

• Generating collision-free motion for robotic manipulators
• Operating in dynamic partially observable environments
• Overcoming limitations of classical and neural planners

Innovation

• Transformer-based neural motion policy pretrained
• Iterative student-teacher finetuning for static avoidance
• Reactive goal-proposal module for dynamic obstacles