# Classical Planning in Deep Latent Space

Source: [Classical Planning in Deep Latent Space](<https://www.sota2.com/research/paper/68fa6cbf1e209131cea35c07>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-28T13:56:53.905Z

Authors: Masataro Asai, Hiroshi Kajino, Alex Fukunaga, Christian Muise

Published: 2021-06-30T21:31:21.000Z

## 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.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Encoder Board Reconstruction on MNIST 8-puzzle Clean](<https://www.sota2.com/research/sota/encoder-board-reconstruction-on-mnist-8-puzzle-clean.md>): Encoder Board Reconstruction; MNIST 8-puzzle Clean; F1 Score: 74.

- [Encoder Board Reconstruction on MNIST 8-puzzle (Noisy)](<https://www.sota2.com/research/sota/encoder-board-reconstruction-on-mnist-8-puzzle-noisy.md>): Encoder Board Reconstruction; MNIST 8-puzzle (Noisy); F1 Score: 70.

- [Encoder Board Reconstruction on IceSlider Clean](<https://www.sota2.com/research/sota/encoder-board-reconstruction-on-iceslider-clean.md>): Encoder Board Reconstruction; IceSlider Clean; F1 Score: 93.

- [Imagination Board Reconstruction on MNIST 8-puzzle Clean](<https://www.sota2.com/research/sota/imagination-board-reconstruction-on-mnist-8-puzzle-clean.md>): Imagination Board Reconstruction; MNIST 8-puzzle Clean; F1 Score: 61.

- [Imagination Board Reconstruction on MNIST 8-puzzle Noisy](<https://www.sota2.com/research/sota/imagination-board-reconstruction-on-mnist-8-puzzle-noisy.md>): Imagination Board Reconstruction; MNIST 8-puzzle (Noisy); F1 Score: 58.

- [Encoder Board Reconstruction on IceSlider Noisy](<https://www.sota2.com/research/sota/encoder-board-reconstruction-on-iceslider-noisy.md>): Encoder Board Reconstruction; IceSlider Noisy; F1-Score: 70.

- [Imagination Board Reconstruction on IceSlider Clean](<https://www.sota2.com/research/sota/imagination-board-reconstruction-on-iceslider-clean.md>): Imagination Board Reconstruction; IceSlider Clean; F1 Score: 62.

- [Imagination Board Reconstruction on IceSlider Noisy](<https://www.sota2.com/research/sota/imagination-board-reconstruction-on-iceslider-noisy.md>): Imagination Board Reconstruction; IceSlider Noisy; F1 Score: 58.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2107.00110>)

[PDF](<https://arxiv.org/pdf/2107.00110.pdf>)