Answers are grounded in this paper and cite their sources.
Background: Parcel lockers consolidate last-mile demand, while vehicle assignment and route sequencing determine the emissions of the resulting distribution network. This study develops a Proximal Policy Optimization (PPO)-based framework for emissions-aware heterogeneous-fleet routing with depot replenishment and explicit multi-trip limits. Methods: Road-network distances and travel times define vehicle-specific speed-dependent emission costs for candidate locker networks with synthetic demand in Tangier, Marrakech, and Casablanca. Twenty independent runs per algorithm on each fixed instance compare the full PPO pipeline with Ant Colony Optimization (ACO), complemented by OR-Tools Guided Local Search, distance-minimizing and emissions-greedy baselines, and component ablations. Results: PPO reduces mean modelled carrier-route emissions relative to ACO by 4.43%, 5.30%, and 3.14%, respectively; all three Welch comparisons remain significant after Holm adjustment. Its mean emissions are also 1.00–1.29% lower than the emissions-oriented OR-Tools benchmark. Demand-perturbation tests demonstrate zero-shot policy reuse on the same networks, and environmental sensitivity experiments preserve the PPO–ACO ranking across the tested scenarios. Online PPO computation is faster than ACO on the two larger instances, with offline training accounted for separately. Conclusions: These findings support the framework’s comparative performance for emissions-aware parcel-locker distribution on the evaluated instances.
No takes yet. Share an insight, caveat, or question.
Ejjanfi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: