Binghong Chen

Abstract. We propose a meta path planning algorithm named 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 exploration and *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.

Illustration of the proposed neurally-guided path planner.

In each epoch, NEXT is executed on a randomly generated planning problem. The search tree grows with \(\tilde V^*\) and \(\tilde \pi^*\) guidance. \(\{\tilde V^*, \tilde\pi^*\}\) will be updated according to the successful path. Such planning and learning iteration is continued interactively.

Short video describing the work. Recorded for ICLR’20.
video link