Reinforcement Learning for Electric Vehicle Traction Motor Control: A Comprehensive Review

The control of traction motors is a key element of speed and torque regulation loops in electric vehicle traction systems, where high efficiency, low torque ripple, and strong robustness to disturbances are required. Conventional control methods show limitations when faced with strong nonlinearities, parametric variations, and unmodeled dynamics. Reinforcement Learning is therefore investigated as a data-driven solution capable of learning optimal control laws without relying on an explicit analytical model of the system. The contribution of the research is a structured and original review dedicated to the application of RL to the control of electric vehicle traction motors. It proposes a systematic classification of algorithms, application domains, and performance objectives. The methodology is based on a bibliographic analysis of works published between 2018 and 2025. RL methods are classified according to the learning paradigm, the control level, and the type of validation. The analyzed algorithms include classical approaches such as Q-learning and SARSA, as well as deep reinforcement learning methods such as DQN, DDPG, and PPO. Control architectures, reward functions, and learning environments are systematically compared. Several studies report superior performance compared to conventional control laws under transient conditions and in the presence of disturbances. Deep learning-based approaches are particularly effective for highly nonlinear systems. However, challenges remain in terms of stability, safety, and computational cost. Experimental validations also confirm the feasibility of real-time implementation within specific hardware constraints. In conclusion, this review highlights the strong potential of RL for traction motor control and outlines perspectives toward safe learning, embedded implementation, and hybrid model-data control strategies.