Guided Game Level Repair via Explainable AI

TL;DR

This paper proposes using explainability methods to identify specific regions of a level that contribute to its unsolvability, by assigning higher weights to these regions, so constraint-based solvers can prioritize these problematic areas, enabling more efficient repairs.

Abstract

Procedurally generated levels created by machine learning models can be unsolvable without further editing. Various methods have been developed to automatically repair these levels by enforcing hard constraints during the post-processing step. However, as levels increase in size, these constraint-based repairs become increasingly slow. This paper proposes using explainability methods to identify specific regions of a level that contribute to its unsolvability. By assigning higher weights to these regions, constraint-based solvers can prioritize these problematic areas, enabling more efficient repairs. Our results, tested across three games, demonstrate that this approach can help to repair procedurally generated levels faster.

Figures (7)

• Figure 1: Visual Representation of the system pipeline.
• Figure 2: Visual representation of different repairs to a Mario level. The left side is the broken level. Each explainability method generates an attribution for all tiles, and these attributions are scaled to weights. The black squares show the low weights marked for the solver. The red squares show the tile that the solver chooses to change. The right side is the repaired level.
• Figure 3: The minimal representation of the levels with 4 tile types (start: {, goal: }, solid: X, and empty: -) used for training the classifiers.
• Figure 4: Representation of how attributions are scaled to meaningful (positive integers) weights for the solver. Lower values indicated a higher priority to change.
• Figure 5: Comparison of number of repaired levels in times between different weight generation methods. The time is shown on a logarithmic scale.