A deep reinforcement learning and fractional packing framework for routing and scheduling in healthcare waste supply chains

Abstract

Artificial intelligence (AI) is increasingly utilized in healthcare logistics, including automated systems for collecting hazardous medical waste from hospitals under strict time and capacity constraints. This study compares three routing algorithms: (1) Google Maps Destination using an application programming interface (API), (2) hybrid clustering with Deep Q-Network (DQN), and (3) a hybrid method combining clustering, the fractional knapsack strategy, and DQN. These algorithms aim to optimize route planning and scheduling for medical waste collection vehicles operating under real-world constraints such as limited vehicle capacity and fixed service windows. The routing problem is modeled as both a capacitated vehicle routing problem (CVRP) and a CVRP with time windows (CVRPTW), capturing complexities. A multi-trip routing strategy is integrated into the promising algorithms to assess its impact on performance metrics, including capacity utilization, travel distance, total operational time, and number of trips. Experimental results indicate hybrid approach with clustering, fractional knapsack, and DQN outperforms others. It achieved capacity utilization rates of 96.47 percent for CVRP and 76.01 % for CVRPTW, requiring six vehicles, a 25 % reduction compared to the Google Maps API method, while matching the performance of clustering with DQN under time constraints. The CVRP model improved capacity utilization by 28.9 % over Google Maps API and 85.1 % over clustering with DQN. Although travel distance increased slightly (0.61 % in CVRP and 7.2 % in CVRPTW), total operational time was reduced by 7.6 and 4.6 %. The proposed model also minimized extra trips, requiring none for CVRP and only one for CVRPTW, compared to two additional trips needed by clustering with DQN in both scenarios. These findings highlight the hybrid approach as a robust, efficient solution for medical waste transportation under complex conditions.