References & Citations
Computer Science > Artificial Intelligence
Title: Classical Planning in Deep Latent Space
(Submitted on 30 Jun 2021 (v1), last revised 15 Jun 2022 (this version, v3))
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.
Submission history
From: Masataro Asai [view email][v1] Wed, 30 Jun 2021 21:31:21 GMT (4936kb,D)
[v2] Tue, 15 Mar 2022 21:14:53 GMT (4471kb,D)
[v3] Wed, 15 Jun 2022 19:00:02 GMT (4480kb,D)
Link back to: arXiv, form interface, contact.