Learning State-Dependent Policy Parametrizations for Dynamic Technician Routing with Rework
研究了家庭维修服务中技师路径规划问题，通过强化学习动态平衡服务紧急度、路径效率和返工风险，发现非理想分配有时能提升整体服务质量。
Home repair and installation services require technicians to visit customers and resolve tasks of different complexities. Technicians often have heterogeneous skills. The geographical spread of customers makes achieving only “ideal” matches between technician skills and task requirements impractical. Additionally, technicians are regularly absent, for example, due to sickness. With only nonideal assignments regarding task requirement and technician skill, some tasks may remain unresolved and require a revisit and rework at a later day, leading to delayed service. For this sequential decision problem, every day, we iteratively build tours by adding “important” customers. The importance bases on analytical considerations and is measured by respecting urgency of service, routing efficiency, and risk of rework in an integrated fashion. We propose a state-dependent balance of these factors via reinforcement learning. We rely on proximal policy optimization (PPO) tailored to the problem specifics, analyzing the implications of specific algorithmic augmentations. A comprehensive study shows that taking a few nonideal assignments can be quite beneficial for the overall service quality. Furthermore, in states where a higher number of technicians are sick and many customers have overdue service deadlines, prioritizing service urgency is crucial. Conversely, in states with fewer sick technicians and fewer customers with overdue deadlines, routing efficiency should take precedence. We further demonstrate the value provided by a state-dependent parametrization via PPO. Funding: This work was supported by the Deutsche Forschungsgemeinschaft [Grants 413322447 and 444657906]. Supplemental Material: The online appendices are available at https://doi.org/10.1287/trsc.2024.0844 .