Deakin University

Exploring Apprenticeship Learning for Player Modelling in Interactive Narratives

This research explores the application of Apprenticeship Learning (AL) to model player behavior within Interactive Narratives (INs).

• Receding Horizon Inverse Reinforcement Learning (RHIRL) can effectively model player behavior in interactive narratives by learning reward functions from human demonstrations to derive imitation policies.
• RHIRL successfully learned action sequences to complete the interactive fiction game Anchorhead, with performance metrics indicating that longer horizons (up to 4 steps) improve convergence to player behavior compared to shorter horizons.
• The study found that configurations prioritizing intention learning—using larger step horizons and assuming sub-optimal player behavior—outperform imitation learning configurations that assume optimal play.