IPPO freight routing beats heuristics with 12.7% higher throughput

A new arXiv paper from researchers at the University of Tennessee (Aliza Sharmin, Xudong Wang, Mustafa Can Camur, Xueping Li) tackles a critical logistics problem: how to route freight when climate extremes simultaneously degrade multiple intermodal corridors. They formulate the problem as a Decentralized Partially Observable Markov Decision Process (Dec-POMDP) with per-shipment action granularity, then train Independent PPO (IPPO) agents under Centralized Training with Decentralized Execution (CTDE). The approach is tested on a 15-hub intermodal network under simulated hurricane disruption, comparing against two heuristic baselines that have privileged state access.

Results across 30 matched episodes show no single policy dominates: IPPO achieves the highest throughput (+12.7%) and delivery rate, while a capacity-aware heuristic leads on Resilience Index (RI) and delay. Under demand surge (2.9:1 capacity ratio), IPPO's RI advantage grows to +6.4%, suggesting learned routing is most valuable when capacity is scarce. Notably, a Multi-Agent PPO (MAPPO) variant with a centralized critic collapses under train-eval queue mismatch (RI=0.811); retraining recovers RI to 1.018 but IPPO still leads on throughput, highlighting residual limitations of centralized critics in per-shipment dispatch settings. The findings offer practical evidence that decentralized multi-agent RL can outperform traditional heuristics for disaster-resilient freight routing, especially under extreme capacity constraints.