Learning to Retrieve Relevant Experiences for Motion Planning

TL;DR

This paper introduces FIRE, a learning-based framework that extracts local representations of motion planning problems and learns a similarity function, enabling more effective retrieval of relevant past experiences to improve motion planning performance.

Contribution

FIRE is the first framework to learn a similarity function over local problem representations, enhancing experience retrieval for motion planning beyond simple hand-crafted methods.

Findings

FIRE outperforms baseline methods in retrieving relevant experiences.

It generalizes well to problems outside its training distribution.

Improves sampling-based planner guidance in complex environments.

Abstract

Recent work has demonstrated that motion planners' performance can be significantly improved by retrieving past experiences from a database. Typically, the experience database is queried for past similar problems using a similarity function defined over the motion planning problems. However, to date, most works rely on simple hand-crafted similarity functions and fail to generalize outside their corresponding training dataset. To address this limitation, we propose (FIRE), a framework that extracts local representations of planning problems and learns a similarity function over them. To generate the training data we introduce a novel self-supervised method that identifies similar and dissimilar pairs of local primitives from past solution paths. With these pairs, a Siamese network is trained with the contrastive loss and the similarity function is realized in the network's latent space. We evaluate FIRE on an 8-DOF manipulator in five categories of motion planning problems with sensed environments. Our experiments show that FIRE retrieves relevant experiences which can informatively guide sampling-based planners even in problems outside its training distribution, outperforming other baselines.