DYNAMIC VEHICLE ROUTING WITH PROMPT CONFIRMATION AND CONTINUALLY OPTIMIZED ROUTE ASSIGNMENTS

Existing dynamic vehicle routing systems for microtransit services either provide prompt confirmation (without continually optimizing vehicle routes, which reduces system efficiency) or continually optimize vehicle routes (without providing immediate service guarantees, which hinders user satisfaction and adoption). The disclosed system overcomes that tradeoff by integrating a prompt confirmation module (that provides near-instantaneous acceptance or rejection) with an anytime optimization module (that continuously improves vehicle manifests during the idle periods between request arrivals). To provide guaranteed service while simultaneously optimizing vehicle route for maximum system efficiency, both the prompt confirmation module and the anytime optimization module are guided by a non-myopic objective function (a state-action value function learned by training a neural network using reinforcement learning and, in some embodiments, accelerated via supervised pre-training) that maximizes the long-term service rate.