Aiming at the problem that driver's participation and driver's state are seldom considered in human-vehicle collaborative steering control, a human-vehicle torque collaborative steering control method based on driver state prediction is proposed. This method takes torque as human-vehicle interface to improve the degree of driver's participation. At the same time, in the process of controller design, taking the driver state into account, the model predictive control method is used, to predict the state of the driver. veDYNA, a high-precision vehicle simulation software, is adopted for simulation verification. The results show that compared with the human-vehicle torque collaborative based on no driver state prediction steering control method, the proposed method can make the auxiliary torque better follow driver's actions, improve vehicle steering performance, and reduce lateral displacement deviation. At the same time, it has good adaptability to different drivers. Furthermore, with the driver's next action as the reference, the current torque can be as close as possible to the desired torque in the next step. Under the condition that the steering performance is almost unaffected, the driver's operating burden can be appropriately reduced.