Value-Based Reinforcement Learning for Task Scheduling in Human-Robot Applications

In modern manufacturing settings, Human-Robot collaboration (HRC) plays a pivotal role in enhancing efficiency, flexibility, and operator well-being. However, existing task allocation algorithms for HRC rely on simplified assumptions and static or heuristic-based approaches. In this paper, we present a value-based Reinforcement Learning (RL) framework for dynamic task scheduling in a collaborative assembly scenario, capable of learning to allocate tasks between human operator and robot in real time, with the goal of minimizing the total tasks completion time. Our presented environment, which can be generalized to other different settings, incorporates both deterministic robot labor and stochastic human behavior. The RL algorithm demonstrates the ability to adaptively manage task assignments under uncertainty, learning efficient policies that generalize across the human variability. This work offers a flexible and scalable foundation for intelligent scheduling in HRC systems and paves the way for future integration of operator-specific parameters and cognitive workload indicators.