VEGA: Electric Vehicle Navigation Agent via Physics-Informed Neural Operator and Proximal Policy Optimization

TL;DR

A vehicle-adaptive energy-aware routing system for electric vehicles (EVs) that integrates physics-informed parameter estimation with RL-based charge-aware path planning and generalizes without retraining to road networks in France and Japan is presented.

Abstract

We present VEGA, a vehicle-adaptive energy-aware routing system for electric vehicles (EVs) that integrates physics-informed parameter estimation with RL-based charge-aware path planning. VEGA consists of two copupled modules: (1) a physics-informed neural operator (PINO) that estimates vehicle-specific physical parameters-drag, rolling resistance, mass, motor and regenerative-braking efficiencies, and auxiliary load-from short windows of onboard speed and acceleration data; (2) a Proximal Policy Optimization (PPO) agent that navigates a charger-annotated road graph, jointly selecting routes and charging stops under state-of-charge constraints. The agent is initialized via behavior cloning from an A* teacher and fine-tuned with cirriculum-guided PPO on the full U.S. highway network with Tesla Supercharger locations. On a cross-country San Francisco-to-New York route (~4,860km), VEGA produces a feasible 20-stop plan with 56.12h total trip time and minimum SoC 11.41%. Against the controlled Energy-aware A* baseline, the distance and driving-time gaps are small (-8.49km and +0.37h), while inference is >20x faster. The learned policy generalizes without retraining to road networks in France and Japan.

Figures (5)

• Figure 1: VEGA overview. A PINO estimates vehicle-specific parameters from speed and acceleration logs, and a PPO planner uses the resulting physics-based energy model to make charge-aware routing decisions on a charger-annotated road graph.
• Figure 2: VEGA system architecture. A PINO module estimates vehicle-specific physical parameters from OBD-II speed and acceleration logs. These parameters feed a physics-based energy model that the PPO routing agent uses to plan charge-aware paths on a real road graph.
• Figure 3: Nonlinear charging profile for the Tesla Model 3 Long Range. Charging from 80% to 100% SoC takes approximately as long as 20% to 80%, motivating the 80% charging cap used by VEGA.
• Figure 4: Convergence of PINO-estimated parameters over a 15-minute driving window. Estimates stabilize within approximately 10 minutes across multiple trials, demonstrating reliable real-time parameter inference.
• Figure 5: Representative VEGA route visualizations for the three reported regions.