Nudge steers robot movements away from obstacles without retraining
Zero-Shot Reactive Obstacle Avoidance for Generative Robot Policies
Robotics
Summary
Robots sometimes bump into things because their movement plans don’t always account for obstacles. The authors propose a new way called NUDGE that helps any robot control system adjust its motions on the fly to avoid crashes, without needing extra training. They use a mathematical method that calculates how close the robot is to obstacles and pushes its planned movement away from danger in real time. This keeps the robot’s original task plan intact while making it safer.
What this means in practice
- •For robotics engineers: Implement real-time obstacle avoidance on robots with diffusion-based policies without retraining the neural networks.
- •For industrial automation teams: Enhance robotic arms to adjust motions reactively to avoid collisions during precise assembly or manipulation tasks.
Authors
Weihang Guo, Lydia E. Kavraki
Abstract
We propose NUDGE (Nudge Update via Differentiable GEometry), a training-free obstacle-avoidance procedure that can be incorporated in any robot policy based on diffusion or flow matching, including diffusion policies and vision-language-action models. Our work injects gradients from a signed distance field, a function returning each point's distance to the nearest obstacle, into the policy at inference time to steer it away from obstacles. It supports any common action parameterization, from absolute or relative joint poses to end-effector poses, through a differentiable joint-trajectory decoder. Experiments show that NUDGE preserves the policy's task distribution and runs reactively in real time.