Anticipatory Monte Carlo tree search–based optimization for stochastic dynamic routing with time windows

Abstract

This paper develops an anticipatory logistics optimization framework for non-profit food rescue operations to address the challenges of hunger and food waste. The study aims to distribute perishable surplus food from food banks to food-insecure households, taking into account uncertain volunteer availability, dynamic household requests, and limited transportation resources. The problem is formulated as a dynamic vehicle routing problem incorporating time windows. A Monte Carlo tree search (MCTS)-based approach is proposed that incorporates vehicle returns to depots for loading food packages. The framework utilizes stochastic rollouts to anticipate future customer arrivals and inform online routing and replenishment decisions. The numerical results indicate that the proposed MCTS framework can effectively solve the problem, outperforming conventional insertion heuristics. Compared to baseline heuristics, the proposed method achieves a 10–15% reduction in total routing cost while serving a larger number of newly arriving household requests under uncertainty.