AI breakthrough could revolutionize how electric vehicle fleets operate

Researchers at Chalmers University of Technology have developed a new method for managing electric vehicle fleets using deep reinforcement learning. The approach tackles the Electric Dial-a-Ride Problem (E-DARP), aiming to optimise routes, charging schedules, and passenger service quality at the same time. So far, the technique remains in the research phase, with testing limited to simulations and ride-sharing data from San Francisco.

The team's method uses a graph neural network encoder paired with an attention-driven policy to solve complex routing challenges. Unlike traditional optimisation techniques, it learns directly from road network features, allowing for faster and more adaptable solutions. Tests on standard benchmark problems showed results within 0.4% of the best-known outcomes while cutting computation times significantly.

Further experiments included sensitivity analyses and robustness checks to measure how changes in key parameters affected performance. These confirmed the policy's reliability under unpredictable, real-world conditions. Evaluations also extended to large-scale scenarios, using San Francisco ride-sharing data to demonstrate scalability.

The breakthrough offers a scalable alternative to conventional methods, which often struggle with the demands of electric cars. By jointly optimising routing, charging, and service quality, the approach addresses the growing need for efficient, on-demand electric transport. However, there is no evidence yet that manufacturers or pilot projects have adopted the system in practice.

The research provides a foundation for smarter management of electric vehicle fleets, balancing energy use, operational costs, and passenger requirements. While still in the testing phase, the method could eventually support real-world applications in uber ride services. Future steps would require practical trials to validate its effectiveness beyond simulations.