**Title: Multi-Objective Reinforcement Learning for Tactical Decision Making for Trucks in Highway Traffic

Abstract:

Ensuring a balance among safety, operational expenses, and efficiency presents a complex decision-making hurdle for heavy-duty vehicles navigating highways. A primary obstacle is that standard scalar reward models, which typically aggregate these competing goals, tend to mask the underlying structure of the necessary trade-offs. To address this, we introduce a multi-objective reinforcement learning framework grounded in Proximal Policy Optimization. This approach is designed to learn a collection of policies that explicitly model these trade-offs and is validated on a scalable simulation platform tailored for the tactical decision-making of trucks.

The method identifies a set of Pareto-optimal policies that delineate the compromises between three conflicting aims: safety (measured by collision rates and successful task completion), energy efficiency (assessed via energy costs), and time efficiency (evaluated through driver costs). The resulting Pareto frontier is both smooth and interpretable, offering the flexibility to select driving behaviors that prioritize various conflicting objectives. Furthermore, this framework facilitates smooth transitions between distinct driving policies without the need for retraining, thereby providing a robust and adaptive decision-making strategy suitable for autonomous trucking systems.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC