Learning to Route Electric Trucks Under Operational Uncertainty

Stavros Orfanoudakis, Ziyan Li, Ruixiao Yang, Nikolay Aristov, Pedro P. Vergara, Chuchu Fan, Elenna Dugundji · 2026

Read Learning to Route Electric Trucks Under Operational Uncertainty by Stavros Orfanoudakis, Ziyan Li, Ruixiao Yang, Nikolay Aristov, Pedro P. Vergara, Chuchu Fan, Elenna Dugundji (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-26566.txt.

Introduction

Orfanoudakis and six colleagues built a learning-based router for electric trucks under battery, charging, and traffic uncertainty. Truck routing couples logistics with energy: limited range, slow charging, and shared chargers defeat heuristics at scale. Their reinforcement-learning formulation uses an event-driven semi-Markov decision model with shared charging, stochastic travel and energy needs, and realistic nonlinear fast charging. A graph state representation plus a rule-based action mask keeps policies admissible. Tests across fleet sizes beat baselines and neared optimization benchmarks while holding high success rates under congestion.

Catalog record

CREATORStavros Orfanoudakis, Ziyan Li, Ruixiao Yang, Nikolay Aristov, Pedro P. Vergara, Chuchu Fan, Elenna Dugundji
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE85462
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.26566

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Learning to Route Electric Trucks Under Operational Uncertainty Stavros Orfanoudakis, Ziyan Li, Ruixiao Yang, Nikolay Aristov, Pedro P. Vergara, Chuchu Fan, Elenna Dugundji Abstract Electric truck operations require routing decisions that remain feasible under limited battery range, long charging times, travel and energy consumption, and competition for shared charging infrastructure. These features make electric truck routing a coupled logistics and energy problem, limiting the practicality of heuristics-based methods and rendering them computationally infeasible at scale. This paper proposes a learning-based framework for the stochastic electric truck routing under charging constraints and operational uncertainty. The problem, solved by Reinforcement Learning, is formulated as an event-driven semi-Markov decision process with shared charging resources, stochastic travel and energy requirements, and realistic nonlinear fast-charging behavior. To support learning in this setting, a graph-based representation of system state and feasible decisions is introduced, together with a rule-based action mask that restricts policies to operationally admissible actions; thus, improving training efficiency. Building on this formulation, an event-driven simulation environment is developed that supports both Reinforcement Learning and benchmarking against heuristic and mathematical programming baselines. Computational experiments across a range of fleet sizes show that the proposed learning-based algorithm consistently outperforms baselines and attains performance close to optimization benchmarks in many settings, while preserving high success rates under charging congestion and uncertainty. [orcid=0000-0002-0767-9488] [orcid=0009-0009-5226-0430] [orcid] [orcid=0009-0007-8832-1907] [orcid=0000-0003-0852-0169] [orcid=0000-0003-4671-233X] [orcid=0000-0001-9277-9485] keywords Reinforcement Learning ,Electric Truck Routing ,Freight Transportation ,Graph Neural Networks ,Stochastic Optimization ,Vehicle Routing † † corresponding: Corresponding author email: [email protected] † † credit: Conceptualization, Methodology, Software, Writing - Original Draft, Writing - Review & Editing † † credit: Conceptualization, Software, Writing - Review & Editing † † credit: Conceptualization, Writing - Original Draft, Writing - Review & Editing † † credit: Conceptualization, Software, Writing - Review & Editing † † credit: Supervision, Funding acquisition, Writing - Review & Editing † † credit: Supervision, Writing - Review & Editing † † credit: Conceptualization, Supervision, Writing - Review & Editing 1 Introduction The electrification of transportation is transforming mobility and logistics systems by creating new opportunities to reduce emissions, improve energy efficiency, and lower dependence on fossil fuels [ 46 ] . Within freight transportation, electric trucks are particularly promising because they can reduce operating costs and local environmental impacts while supporting the decarbonization of regional and urban logistics. At the same time, their deployment changes the nature of routing and dispatching decisions [ 18 ] . Unlike conventional trucks, electric trucks operate under limited driving range, long charging times, and uneven charging infrastructure, making charging a central component of route planning rather than a secondary operational consideration. As a result, electric truck routing becomes a coupled logistics and energy management problem in which route feasibility and service performance depend not only on delivery sequences [ 35 ] , but also on when, where, and how vehicles charge, especially in fleet settings with shared and capacity-constrained charging resources [ 5 ] . This coupling becomes especially consequential in operational settings where routing decisions must be made dynamically and under uncertainty. In practice, electric truck operations are affected not only by travel distances and customer sequences, but also by charging timing, location, and duration [ 36 ] . Public or shared charging stations may be capacity-constrained, subject to congestion, and used by multiple stakeholders, so a truck’s feasible route depends in part on the behavior of other actors in the system. As a result, route feasibility and service performance are shaped jointly by battery state, charging availability, queueing delays, and stochastic travel and service conditions [ 14 ] . These features are amplified at the fleet level, where several trucks may compete for limited charging capacity [ 28 ] and where charging delays are endogenous rather than exogenously specified [ 44 ] . In such environments, methods for electric truck routing must do more than identify short paths, they must also support operationally feasible, time-sensitive decisions in systems where routing and charging are tightly interdependent. The electric truck fleet routing problem (eTFRP) [ 21 ] has evolved rapidly from a niche extension of the traditional vehicle routing problem (VRP) and electric VRP (eVRP) into a distinct research stream centered on the operational realities of electrified freight systems [ 7 ] . Optimization-based methods provide the main analytical foundation by embedding charging decisions, uncertainty, and energy-feasibility constraints directly into mathematical models [ 21 ] . This line includes stochastic formulations for routing with simplified charging curves [ 36 ] , robust models for heavy-duty electric trucks under uncertain energy consumption [ 2 ] , and conic mixed-integer programs that capture richer physical drivers of energy use such as road conditions, vehicle dynamics, payload, and distance [ 43 ] . These approaches offer strong interpretability and disciplined feasibility control, but their computational requirement grows rapidly with fleet size, uncertainty, and modeling realism, often requiring decomposition or approximation to remain tractable [ 19 ] . Metaheuristics offer a more scalable alternative by combining charging-aware route construction, local improvement, and repair mechanisms within flexible search procedures. Variable-neighborhood search [ 10 ] , memetic [ 9 ] , and ant-colony-based methods [ 13 ] have shown that large instances can be handled more efficiently in practice, although their performance often depends on instance structure and parameter tuning and is harder to characterize systematically under uncertainty. In parallel, Reinforcement Learning (RL) has emerged as a promising direction [ 30 ] because it enables search across instances during training and enables rapid inference at deployment time [ 17 ] . For classical routing problems, attention-based and transformer-style policies, together with improved rollout and symmetry-based training strategies, have demonstrated that learned policies can achieve strong performance without hand-crafted search rules [ 45 , 15 ] . Recent studies have extended these ideas to charging-aware, stochastic, and safety-aware electric vehicle routing settings [ 4 ] . However, much of this literature still remains closer to simple eVRP settings than to more realistic fleet-level electric truck operations with shared charging resources, traffic congestion, and truck-specific operational constraints [ 23 , 38 ] . In addition, learning-based approaches rarely embed explicit feasibility control in the decision process, and comparisons across optimization, heuristic, and learning methods are often difficult to interpret because they rely on different assumptions about uncertainty, charging behavior, and temporal dynamics [ 22 ] . These gaps motivate the need for a truck-oriented, learning-compatible framework that combines a realistic operational structure, feasibility-aware decision-making, and systematic benchmarking within a common evaluation pipeline. These gaps point to a clear need for methods that retain the operational realism of electric truck routing while remaining scalable enough for repeated, time-sensitive decision making. In response, this paper develops a truck-oriented and learning-compatible framework 1 1 1 The project repository is available at https://github.com/StavrosOrf/eTruckRouting . for the eTFRP under charging constraints and uncertainty. The problem is formulated to explicitly capture fleet-level interactions through shared charging infrastructure, stochastic travel and energy conditions, and realistic charging behavior, while preserving the structure needed for sequential decision-making. On this basis, Graph Proximal Policy Optimization (GraphPPO) is introduced as a graph neural network (GNN)-based [ 16 ] RL approach that combines a structured state–action representation with an explicit state-based feasibility action mask, allowing routing and charging decisions to be

Read the complete text