Exploring reinforcement learning in path planning for omnidirectional robot soccer

Abstract

Path Planning consists of a widely studied computational problem of great applicability in autonomous robotics and virtual reality environments that aims to solve the following problem: given the origin of an entity in space, obtain a feasible collision-free route to the destination. From the characteristics of a given environment, in this case, a soccer field in conventional game conditions that imply a greater complexity given the dynamics of the obstacles, it is intended to use Reinforcement Learning – technique that has gained expression over time due to the ability of its applications to perform better than humans in different scenarios –, to optimize the trajectories performed by an agent as it maximizes the reward accumulated within the executed iterations.