Dynamic Player Modelling in Serious Games Applied to Rehabilitation Robotics

摘要

This article proposes a reinforcement learning approach to dynamically model the player skills in applications that integrate games and rehabilitation robotic. The approach aims to match the game difficulty to the player skills, keeping proper motivation (flow) during a rehabilitation process. The traditional rehabilitation process involves repetitive exercises. Robots and serious games provide new means to improve user motivation and commitment during treatment. Each person shows different skills when facing the challenges posed by computer games. Thus, the game difficulty level should be adjusted to each player skill level. The Q-Learning algorithm was adapted in this context to modify game parameters and to assess user skills based on a performance function. This function provides a path to an individual difficulty adjustment and consequently a tool to keep the user exercising. Experiments with thirty minutes duration are presented, involving four players, and the results obtained indicate the proposed approach is feasible for modeling the user behaviour getting to capture the adaptations and trends for each player according to the game difficulties.