Classical Planning in Deep Latent Space

TL;DR

Latplan is an unsupervised deep learning architecture that automatically learns symbolic planning models from images, enabling classical planning without manual knowledge encoding.

Contribution

It introduces Latplan, which combines deep learning with classical planning to learn symbolic models from unlabeled image data, bypassing the knowledge acquisition bottleneck.

Findings

Successfully learned planning models for 6 domains from images

Generated plans in visual domains like 8-puzzle and Sokoban

Demonstrated the feasibility of unsupervised symbolic model learning

Abstract

Current domain-independent, classical planners require symbolic models of the problem domain and instance as input, resulting in a knowledge acquisition bottleneck. Meanwhile, although deep learning has achieved significant success in many fields, the knowledge is encoded in a subsymbolic representation which is incompatible with symbolic systems such as planners. We propose Latplan, an unsupervised architecture combining deep learning and classical planning. Given only an unlabeled set of image pairs showing a subset of transitions allowed in the environment (training inputs), Latplan learns a complete propositional PDDL action model of the environment. Later, when a pair of images representing the initial and the goal states (planning inputs) is given, Latplan finds a plan to the goal state in a symbolic latent space and returns a visualized plan execution. We evaluate Latplan using image-based versions of 6 planning domains: 8-puzzle, 15-Puzzle, Blocksworld, Sokoban and Two variations of LightsOut.