Learning State-Dependent Policy Parametrizations for Dynamic Technician Routing with Rework

• DOI: 10.1287/trsc.2024.0844
• Publication Date: 2025-05-15T17:03:26Z
• Authors: Jonas Stein; Florentin D. Hild...; Marlin W. Ulmer; Barrett W. Thomas
• Keywords: Technician

Home repair and installation services require technicians to visit customers resolve tasks of different complexities. Technicians often have heterogeneous skills. The geographical spread makes achieving only “ideal” matches between technician skills task requirements impractical. Additionally, are regularly absent, for example, due sickness. With nonideal assignments regarding requirement skill, some may remain unresolved a revisit rework at later day, leading delayed service. For this sequential decision problem, every we iteratively build tours by adding “important” customers. importance bases on analytical considerations is measured respecting urgency service, routing efficiency, risk in an integrated fashion. We propose state-dependent balance these factors via reinforcement learning. rely proximal policy optimization (PPO) tailored the problem specifics, analyzing implications specific algorithmic augmentations. A comprehensive study shows that taking few can be quite beneficial overall service quality. Furthermore, states where higher number sick many overdue deadlines, prioritizing crucial. Conversely, with fewer efficiency should take precedence. further demonstrate value provided parametrization PPO. Funding: This work was supported Deutsche Forschungsgemeinschaft [Grants 413322447 444657906]. Supplemental Material: online appendices available https://doi.org/10.1287/trsc.2024.0844 .