Dynamic difficulty adjustment with Evolutionary Algorithm in games for rehabilitation robotics

Abstract

This article explores game difficulty adjustment for serious game applications in rehabilitation robotics. In this context, a difficulty adjustment system is proposed that takes user performance as input and generates two different responses: a) a change in the distance the user should cover, and b) the velocity provided to the target. User performance is estimated from its ability to achieve the targets (game score) performing movements. The system interference in user displacement value and target speed where chosen to stimulate the user to achieve specific rehabilitation goals. The game difficulty adjustment has received small attention in the context of rehabilitation robotics interfaces. It is important to note that games developed for rehabilitation differ from commercial entertainment games due to severe limitations imposed to patients by pathologies like stroke, cerebral palsy and spinal cord injury. An Evolutionary Algorithm (AE) based optimization strategy was adopted to adjust game's difficulty. A meta-profile for user behavior was also developed allowing to create and simulate different virtual users and game experiences in computer. This user profile includes a reaction time (time delay), motion disturbance and a kinematical motion profile based on a polynomial function. Using the meta-profile, different user motion behavior can be generated for exhaustive test and optimization of the difficulty adjustment system. The approach allows the reduction of development time and also the reduction in the number of experiments with volunteers. The computer simulation test results are presented to demonstrate the capacity of the difficulty adjustment system to adapt the game characteristics to the users' abilities with different skills levels.