Genetic Fuzzy System for Anticipating Athlete Decision Making in Virtual Reality

Abstract

Intercepting and impeding an opponent is a fairly common behavior in contact and collision sports such as soccer, football, lacrosse or basketball. In soccer, for example, the main objective of a defender is to intercept and impede an attacking opponent as he or she navigates toward the goal. These athlete vs. athlete interactions often lead to collisions, and the uncertainty surrounding them frequently leads to injury. A virtual reality (VR) training platform with non-player characters (NPC) that can anticipate an athletes decisions would, therefore, be a desirable tool to be used by sports trainers to safely and effectively promote the resiliency of athletes to these types of situations. Here we applied this platform to a VR task that required the athletes to run past a series of NPCs to reach a stationary virtual waypoint, or goal. Each NPC is modeled as a Genetic Fuzzy System (GFS) that is trained using a new methodology, called FuzzyBolt, that is capable of training large fuzzy logic systems efficiently to provide better predictive quality. The end result is that such an intelligent NPC is able to more accurately predict athlete movements such that it becomes more difficult for the athlete to successfully navigate around the NPC and to the virtual goal. This, in turn, forces the athlete to develop new movement and decision making strategies in order to evade the NPC, thus enhancing their resiliency and ultimately reducing the risk of collision-based injury on the field of play.