Abstract: Human-Machine interaction has become one of the focuses of intelligent vehicle design. Aiming at the problem of human-machine shared steering control, a human-machine interaction method with variable impedance based on reinforcement learning is proposed. Firstly, A human-machine interaction framework for shared steering is built based on virtual impedance, which can describe the continuous process of control authority distribution. Secondly, on this basis, a variable impedance-based human-machine shared steering control method is designed, which can dynamically distribute control authority by changing virtual impedance. Thirdly, an impedance tuning strategy based on deep deterministic policy gradient (DDPG) is developed to determine the virtual impedance according to the driver's steering behavior. The driver-in-the-loop experiment shows that, compared with the conventional method, the method proposed can make the automation system yield a certain degree of control authority to the driver according to the driver’s steering behavior, and make the interaction process smooth and easy for the driver to adapt to. The method has a less bad influence on the driver and reduces the driver's steering load, and meanwhile, the automation system can also generate an appropriate control torque to express its intention to the driver, and realize the effective human-machine interaction.

Key words: intelligent vehicle, human-machine interaction, shared steering, impedance control, reinforcement learning