AI generates enemy body shapes from player collision data
Researchers create enemies that learn from how players bump into them
Deep Dive
Researchers Gonzalez and Guzdial explore three novel approaches to procedurally generate enemy morphologies based on player collision information. Each method has distinct strengths and weaknesses, and all perform equivalently or better than an evolutionary baseline adapted from robotics. This work addresses a gap in procedural content generation, where generating enemy body plans has seen little prior research.
Key Points
- Three novel collision-based methods for generating enemy morphologies, using heatmaps, gradients, and hybrids
- Outperforms or matches evolutionary baseline adapted from robotics morphology generation
- First systematic work on generating enemy body plans using player interaction data in games
Why It Matters
Enables adaptive enemy design that evolves with player behavior, boosting replayability and personalizing difficulty.