Graph Neural Networks for Motion Planning

About

This paper investigates the feasibility of using Graph Neural Networks (GNNs) for classical motion planning problems. We propose guiding both continuous and discrete planning algorithms using GNNs' ability to robustly encode the topology of the planning space using a property called permutation invariance. We present two techniques, GNNs over dense fixed graphs for low-dimensional problems and sampling-based GNNs for high-dimensional problems. We examine the ability of a GNN to tackle planning problems such as identifying critical nodes or learning the sampling distribution in Rapidly-exploring Random Trees (RRT). Experiments with critical sampling, a pendulum and a six DoF robot arm show GNNs improve on traditional analytic methods as well as learning approaches using fully-connected or convolutional neural networks.

Arbaaz Khan, Alejandro Ribeiro, Vijay Kumar, Anthony G. Francis · 2020

Related benchmarks

Task	Dataset	Result
2D Navigation Value Function Prediction	Disk ID (test)	L2 Error16.3	7
2D Navigation Value Function Prediction	HouseExpo ID (test)	L2 Error19.1	7
2D Navigation Value Function Prediction	City Streets ID (test)	L2 Error9.4	7
2D Navigation Value Function Prediction	Square OOD (test)	L2 Error3.9	7
2D Navigation Value Function Prediction	City Streets OOD (test)	L2 Error23.2	7
Value Function Learning (Corridor Objective)	Disk ID (test)	Value Error L2 (%)21.3	7
Value Function Learning (Corridor Objective)	Disk OOD (test)	Value Function L2 Error (%)24.9	7
2D Navigation Value Function Prediction	Square ID (test)	L2 Error3	7
2D Navigation Value Function Prediction	Maze ID (test)	L2 Error39.4	7
2D Navigation Value Function Prediction	Disk OOD (test)	L2 Error18.7	7