What determines difficulty of solving a problem? Although this question has been studied before, we found examples which show large differences in problem difficulty which are not explained by concepts identified in previous research. This differences are caused mainly by the structure of a problems’ state spaces and cannot be easily captured by static metrics like size of the state space or th...

In this work, we develop a Monte Carlo Tree Search based approach to procedurally generate Sokoban puzzles. To this end, we propose heuristic metrics based on surrounding box path congestion and level tile layout to guide the search towards interesting puzzles. Our method generates puzzles through simulated game play, guaranteeing solvability in all generated puzzles. Our algorithm is efficient...

The objective of this experiment was to use multiple Flockbots to solve sokoban puzzles in parallel (a multibody simulation with a central planner). The simulation takes place an 8x7 grid that we constructed. Each grid space has a QR code in it for localization. There are three main components to solving the problem: the centralized planner, which finds a solution to the puzzle and coordinates ...

In this work, we develop a Monte Carlo Tree Search based approach to procedurally generate Sokoban puzzles. To this end, we propose heuristic metrics based on surrounding box path congestion and level tile layout to guide the search towards interesting puzzles. Our method generates puzzles through simulated game play, guaranteeing solvability in all generated puzzles. Our algorithm is efficient...

We show that the optimization problem is NP-hard for a wide class of motion planning puzzles, including classical SOKOBAN. We investigate a new problem, the Box Mover Problem (BMP), in which the agent is allowed to lift and carry boxes on a rectilinear grid in order to rearrange them. Some classical motion planning puzzles are special cases of BMP. We also identify a natural class of BMP instan...

Deep learning for procedural level generation has been explored in many recent works, however, experimental comparisons with previous works are rare and usually limited to the work they extend upon. The goal of this article is conduct an study on four deep generators Sokoban explore their stren...

Humans can construct powerful mental programs for many domains never seen before. We address the questions of how this occurs, and how it could possibly be accomplished in software. Section one surveys a theory of natural understanding, as follows. One understands a domain when one has mental programs that can be executed to solve problems arising in the domain. Evolution created compact progra...

This paper introduces RL Brush, a level-editing tool for tile-based games designed mixed-initiative co-creation. The uses reinforcement-learning-based models to augment manual human level-design through the addition of AI-generated suggestions. Here, we apply Brush designing levels classic puzzle game Sokoban. We put online and tested it in 39 different sessions. results show that users using A...