Action Schema Networks: Generalised Policies with Deep Learning

TL;DR

This paper presents Action Schema Networks (ASNet), a neural network architecture that learns generalized policies for probabilistic planning by leveraging relational structure and weight sharing, enabling efficient training across problem domains.

Contribution

Introduces ASNet, a neural network architecture that mimics planning problem structure, allowing for domain-generalized policy learning with efficient training and robustness.

Findings

ASNet outperforms traditional non-learning planners in multiple challenging domains.

The weight-sharing scheme enables application to any problem within a planning domain.

Training on small problems generalizes well to larger problems.

Abstract

In this paper, we introduce the Action Schema Network (ASNet): a neural network architecture for learning generalised policies for probabilistic planning problems. By mimicking the relational structure of planning problems, ASNets are able to adopt a weight-sharing scheme which allows the network to be applied to any problem from a given planning domain. This allows the cost of training the network to be amortised over all problems in that domain. Further, we propose a training method which balances exploration and supervised training on small problems to produce a policy which remains robust when evaluated on larger problems. In experiments, we show that ASNet's learning capability allows it to significantly outperform traditional non-learning planners in several challenging domains.