Modulating Game Difficulty with Machine Learning
summary
Prototype pairs Unreal Engine Behavior Trees and Unreal's Environment Query System with a neural network that predicts time-to-capture from snapshots of positions, directions and orb states, then scores EQS candidate patrol points to pick easy/medium/hard locations. Data comes from 400 simulated runs, predictions correlate with subjective difficulty and produce a clear difficulty gradient; hard ML-driven EQS matches an aggressive hand‑crafted EQS in kill rate. EQS-driven patrol selection, ML predicts time-to-capture, three thresholded difficulty modes.
AI-written summary · about
chapters (8)
- 0:00Introduction and overview
- 1:02Pac-Man AI analysis
- 6:12Machine learning for difficulty
- 9:31Prototype methodology
- 13:32Implementing EQS and ML
- 20:09Data training and results
- 28:23Evaluating difficulty modes
- 35:31Future work and conclusions
related talks
03mExploration of Player Emotions, Behaviours, and Individual Differences across Game Difficulty...
59mGame AI 4N00bz
48mMachine Takeover - Improving Your Game With Machine Learning & AI
24mEngaging new and diverse audiences with ML
53mBalancing gameplay of aRPG using player behavioral patterns
30mLeveraging Sentiment and Topic Detection Analysis for Decoding Player..
20mWhat can AI do for me?: evaluating machine learning interpretations in cooperative play
59mExtended Machine Learning