Unsupervised Path Regression Networks

TL;DR

This paper introduces an unsupervised neural network approach for solving shortest path problems by directly regressing spline paths, outperforming supervised methods in scalability and speed without needing ground truth paths.

Contribution

It presents a novel unsupervised training method using a geometry-dependent cost function for collision-free path regression, improving over existing supervised techniques.

Findings

Outperforms state-of-the-art supervised learning baselines.

Offers a more scalable training pipeline.

Achieves significant inference speedup.

Abstract

We demonstrate that challenging shortest path problems can be solved via direct spline regression from a neural network, trained in an unsupervised manner (i.e. without requiring ground truth optimal paths for training). To achieve this, we derive a geometry-dependent optimal cost function whose minima guarantees collision-free solutions. Our method beats state-of-the-art supervised learning baselines for shortest path planning, with a much more scalable training pipeline, and a significant speedup in inference time.