Management of Braking Energy in Electric Vehicles using Reinforcement Learning

The uncertainty and nonlinearity of processes in electrically driven vehicles complicate the use of traditional controllers in braking systems. In addition to the variety of existing algorithms designed for specific deceleration conditions, this study proposes a reinforcement learning-based methodology that can adapt to changing environmental characteristics. As soon as the brake pedal is pressed, the required vehicle and road parameters are transmitted to a neural network that generates output data by means of deep learning. The presented system makes a decision regarding the strength of the braking torque and its distribution between electrical and friction brakes with regard to the current situation. To verify the quality of the offered approach, simulation was conducted, and the results show that the new system ensures significant energy recovery, thereby promising an increase of braking efficiency.