Motion Policy Networks

TL;DR

Motion Policy Networks (MπNets) are an end-to-end neural model that rapidly generates collision-free, smooth robot motions from a single depth camera, outperforming prior methods in speed, robustness, and transferability to real robots.

Contribution

This paper introduces MπNets, a novel neural motion planning model trained on extensive simulation data, achieving real-time, reliable, and transferable collision-free motion generation.

Findings

MπNets are 46% better than prior neural planners.

They are significantly faster than global planners.

MπNets transfer well to real robots despite training only in simulation.

Abstract

Collision-free motion generation in unknown environments is a core building block for robot manipulation. Generating such motions is challenging due to multiple objectives; not only should the solutions be optimal, the motion generator itself must be fast enough for real-time performance and reliable enough for practical deployment. A wide variety of methods have been proposed ranging from local controllers to global planners, often being combined to offset their shortcomings. We present an end-to-end neural model called Motion Policy Networks (M$\pi$Nets) to generate collision-free, smooth motion from just a single depth camera observation. M$\pi$Nets are trained on over 3 million motion planning problems in over 500,000 environments. Our experiments show that M$\pi$Nets are significantly faster than global planners while exhibiting the reactivity needed to deal with dynamic scenes. They are 46% better than prior neural planners and more robust than local control policies. Despite being only trained in simulation, M$\pi$Nets transfer well to the real robot with noisy partial point clouds. Code and data are publicly available at https://mpinets.github.io.