Learning to Plan in High Dimensions via Neural Exploration-Exploitation Trees

TL;DR

NEXT is a neural meta-planning algorithm that leverages prior experience and learned search directions to efficiently solve high-dimensional path planning problems, outperforming classical methods.

Contribution

The paper introduces Neural Exploration-Exploitation Trees (NEXT), a novel neural architecture that learns promising search directions and integrates them into a UCB-based planner for high-dimensional spaces.

Findings

NEXT achieves higher sample efficiency than RRT in high dimensions.

NEXT produces more compact search trees compared to existing methods.

Experimental results show NEXT significantly outperforms state-of-the-art planners on benchmarks.

Abstract

We propose a meta path planning algorithm named \emph{Neural Exploration-Exploitation Trees~(NEXT)} for learning from prior experience for solving new path planning problems in high dimensional continuous state and action spaces. Compared to more classical sampling-based methods like RRT, our approach achieves much better sample efficiency in high-dimensions and can benefit from prior experience of planning in similar environments. More specifically, NEXT exploits a novel neural architecture which can learn promising search directions from problem structures. The learned prior is then integrated into a UCB-type algorithm to achieve an online balance between \emph{exploration} and \emph{exploitation} when solving a new problem. We conduct thorough experiments to show that NEXT accomplishes new planning problems with more compact search trees and significantly outperforms state-of-the-art methods on several benchmarks.