# Procedural Generation of Initial States of Sokoban

## TL;DR

This paper introduces a system called Beta that uses novel hardness metrics and exploration techniques to generate challenging, solvable Sokoban initial states, surpassing human-designed puzzles in difficulty.

## Contribution

The paper presents new hardness metrics based on pattern database heuristics and a novelty-driven search method for generating difficult Sokoban initial states.

## Key findings

- Beta generates initial states harder than human-designed puzzles.
- Hardness metrics effectively guide the search process.
- Generated states are solvable and suitable for evaluating planning systems.

## Abstract

Procedural generation of initial states of state-space search problems have applications in human and machine learning as well as in the evaluation of planning systems. In this paper we deal with the task of generating hard and solvable initial states of Sokoban puzzles. We propose hardness metrics based on pattern database heuristics and the use of novelty to improve the exploration of search methods in the task of generating initial states. We then present a system called Beta that uses our hardness metrics and novelty to generate initial states. Experiments show that Beta is able to generate initial states that are harder to solve by a specialized solver than those designed by human experts.