Neural MP: A Generalist Neural Motion Planner

TL;DR

This work builds a large number of complex scenes in simulation, collects expert data from a motion planner, then distills it into a reactive generalist policy, and combines this with lightweight optimization to obtain a safe path for real world deployment.

Abstract

The current paradigm for motion planning generates solutions from scratch for every new problem, which consumes significant amounts of time and computational resources. For complex, cluttered scenes, motion planning approaches can often take minutes to produce a solution, while humans are able to accurately and safely reach any goal in seconds by leveraging their prior experience. We seek to do the same by applying data-driven learning at scale to the problem of motion planning. Our approach builds a large number of complex scenes in simulation, collects expert data from a motion planner, then distills it into a reactive generalist policy. We then combine this with lightweight optimization to obtain a safe path for real world deployment. We perform a thorough evaluation of our method on 64 motion planning tasks across four diverse environments with randomized poses, scenes and obstacles, in the real world, demonstrating an improvement of 23%, 17% and 79% motion planning success rate over state of the art sampling, optimization and learning based planning methods. Video results available at mihdalal.github.io/neuralmotionplanner

Figures (10)

• Figure 1: Neural Motion Planning at Scale in the Real World Our approach enables a single, generalist neural network policy to solve motion planning problems across diverse setups; Neural MP can generate collision free motions for a wide array of unseen tasks significantly faster and with higher success than traditional as well as learning-based motion planning approaches.
• Figure 2: Visualization of Diverse Simulation Training Environments: We train Neural MP on a wide array of motion planning problems generated in simulation, with significant pose, procedural asset, and mesh configuration randomization to enable generalization.
• Figure 3: Method Overview: We present Neural Motion Planners, which consists of 3 main components. Left: Large Scale data generation in simulation using expert planners Middle: Training deep network models to perform fast reactive motion planning Right: Test-time optimization at inference time to improve performance.
• Figure 4: Emergent Capabilities of Neural MP
• Figure 5: Test-time Optimization Analysis For the Bins Scene 1 task, we plot the number of points in collision across 100 sampled trajectories from the model. 25% of the trajectories are completely collision free and we select a trajectory execute from that subset.