Reinforcement learning for humanitarian relief distribution with trucks and UAVs under travel time uncertainty

• DOI: 10.1016/j.trc.2023.104401
• Publication Date: 2023-11-07T23:33:55Z
• Authors: Robert van Steenb...; Martijn Mes; Wouter van Heeswijk
• Keywords: Heuristics; Time horizon; Vehicle Routing Problem; Robustness

Effective humanitarian relief operations are challenging in the aftermath of disasters, as trucks often faced with considerable travel time uncertainties due to damaged transportation networks. Efficient deployment Unmanned Aerial Vehicles (UAVs) potentially mitigates this problem, supplementing truck fleets an impactful manner. To plan last-mile distribution setting, we introduce a multi-trip, split-delivery vehicle routing problem and UAVs, soft windows, stochastic times for distribution, formulated dynamic program. Within finite horizon, aim maximize weighted objective function comprising number goods delivered, different locations visited, late arrival penalties. Our study offers insights into dealing uncertainty logistics by (i) deploying partial substitutes trucks, (ii) evaluating solutions generated two deep reinforcement learning (RL) approaches – specifically value approximation (VFA) policy (PFA) (iii) comparing RL stemming from mathematical programming heuristics. Experiments performed on both Solomon-based instances real-world cases. The cases 2015 Nepal earthquake 2018 Indonesia tsunami based locally collected field data UAV specifications, provide practical insights. experimental results show that decision-making improves performance robustness operations, achieving reductions lateness penalties around 85% compared static expected times. Furthermore, replacing half UAVs 11% 56%, benefitting reliability location coverage. indicate use methods successfully mitigate operations.