Introducing Human Corrective Multi-Team SRR Sports Scheduling via Reinforcement Learning

Scheduling athletic matches is a challenging problem where the goal is to create robust timetables that not only are fair for all players, but also satisfy kaleidoscopic constraints. Although many advancements have been made that involve combinatorial optimization and metaheuristic techniques, they are limited to fixed scenarios and ideal conditions. Moreover, these methods do not consider the realistic and dynamic demands on teams during the scheduling process. Motivated by these limitations, this work explores emulating and integrating human corrective learning in reinforcement learning (RL) to automatically perform multi-team single round robin (SRR) scheduling while satisfying minimum breaks. Our schedule-then-break approach builds on first modeling the RL agent's actions as a Markov decision process (MDP) which consists of the agent's HAP assignments as well as its corrective actions to ensure that these constraints are made. Next, a SRR reinforcement learning environment is proposed where one-step temporal difference methods are employed to assess how well the agent schedules and corrects as it produces completed timetables. Our simulative results are promising which further illustrate that the agent is effectively able to perform scheduling assignments while also discerning when to make corrections. The benefit of this approach and analysis is that it can now serve as a foundational framework for realistic sports-type scheduling scenarios particularly for cases when additional constraints are incorporated while intermittent schedules are made.