Tariq Osei-Mensah

Separating opponent behavior from game mechanics unlocks models that transfer across rule changes.

Procedural generation keeps RL agents from overfitting to fixed training environments.

Game policies trained on narrow conditions collapse on unseen levels.