Multi-objective deep reinforcement learning strategy paves the way for safer, greener autonomous electric mobility

The rapid rise of electric vehicles combined with breakthroughs in autonomous driving technology is reshaping the future of transportation toward greater sustainability. Intelligent electric vehicles, particularly plug-in hybrid electric vehicles (PHEVs), hold immense potential to slash energy consumption and curb emissions through smarter, more coordinated control of motion and powertrain operations. Yet achieving this dual goal of uncompromising safety and superior energy efficiency remains challenging. Traditional approaches often treat driving safety, eco-friendly trajectory planning, and powertrain energy management as separate tasks, leading to trade-offs that limit overall performance in complex, real-world driving scenarios.

To bridge this gap, researchers have introduced the Intelligent Eco-Driving Strategy (IEDS), an end-to-end autonomous driving solution powered by refined deep reinforcement learning (DRL). Built around a multi-head deep Q network (DQN), the IEDS processes real-time inputs such as the vehicle’s own state, surrounding traffic environment, and road conditions to simultaneously generate decisions on lane changing, speed adjustment, and torque distribution between the engine and electric motor. A carefully designed multi-objective reward function guides the learning process, penalizing risky maneuvers that could lead to collisions while rewarding smooth, efficient behaviors that minimize fuel use. This integrated approach allows the system to balance potentially conflicting priorities within a single neural network framework, enabling the vehicle to navigate dynamic obstacles while optimizing energy flow without relying on predefined driving cycles or rule-based heuristics.

Extensive simulations demonstrate the effectiveness of the IEDS in high-speed highway scenarios. The strategy consistently delivers stable, safe driving with an average speed around 75 km/h and specific fuel consumption below 240 g/(kW·h), maintaining safe operation for over 160 seconds out of a maximum 180-second evaluation window. When benchmarked against baseline methods, the IEDS improves obstacle avoidance capability by 2.10% and enhances energy-saving performance by 5.83%. Most strikingly, across a wide range of randomized driving conditions, the IEDS achieves 97.07% of the theoretical optimum energy management performance determined by dynamic programming (DP), far surpassing the 76.47% and 91.72% attained by comparative strategies. These gains highlight the system’s ability to produce near-optimal results in real-time while preserving driving safety and comfort, even when trained primarily around a reference speed of 20 m/s and then tested across varying velocities.

The benefits extend beyond laboratory metrics to tangible societal and environmental impacts. By enabling PHEVs to operate closer to their theoretical energy-saving limits without sacrificing collision avoidance, the IEDS can contribute to meaningful reductions in fuel consumption and greenhouse gas emissions across fleets of intelligent vehicles. Its data-driven nature eliminates dependence on hand-crafted rules or specific route knowledge, making it adaptable to diverse traffic patterns and road conditions. This robustness positions the strategy as a practical step toward scalable, green autonomous mobility that aligns with global sustainability goals.

Looking forward, the IEDS opens promising avenues for broader deployment and refinement. Future work could incorporate additional sensor modalities, such as LiDAR or V2X communication, to further enhance perception in adverse weather or low-visibility settings. Extending the framework to urban environments with denser interactions or integrating it with vehicle-to-grid energy systems could unlock new layers of efficiency. Real-vehicle testing and hardware-in-the-loop validation will be essential next steps to confirm transferability from simulation to physical platforms, potentially accelerating the adoption of holistic eco-driving capabilities in next-generation intelligent electric vehicles.

In essence, this research represents a significant leap in unifying safety, eco-driving, and energy management within a cohesive DRL-based controller. By achieving near-optimal energy performance alongside robust obstacle avoidance, the IEDS addresses a critical barrier to the widespread realization of sustainable autonomous mobility. As intelligent electric vehicles become more prevalent, such integrated strategies will play an increasingly vital role in driving down energy demand, cutting emissions, and advancing cleaner, safer transportation systems for the future.

Reference

Author: He Tong ^a, Liang Chu ^a, Zheng Chen ^b, Yonggang Liu ^c, Yuanjian Zhang ^d, Jincheng Hu ^d

Title of original paper: Multi-objective autonomous eco-driving strategy: A pathway to future green mobility

Article link: https://www.sciencedirect.com/science/article/pii/S2773153725000295

Journal: Green Energy and Intelligent Transportation

DOI: 10.1016/j.geits.2025.100279

Affiliations:

^aCollege of Automotive Engineering, Jilin University, Changchun 130022, China

^bFaculty of Transportation Engineering, Kunming University of Science and Technology, Kunming 650500, China

^cState Key Laboratory of Mechanical Transmissions & College of Mechanical and Vehicle Engineering, Chongqing University, Chongqing 400044, China

^dDepartment of Aeronautical and Automotive Engineering, Loughborough University, Leicestershire, LE11 3TU, United Kingdom