Human–machine collaborative (HMC) torque control allows the drivers to participate in driving and cooperate to complete driving tasks. However, the driver's torque input and unknown disturbances encountered by the vehicle will adversely affect the control of autonomous vehicles. The problems of "explosion of complexity" appear when the adaptive backstepping method is used to design the steering control strategy. To solve the above problems, a steering control strategy based on command filter adaptive torque control is proposed. Firstly, in order to describe the dynamic response of the vehicle and the maneuvering behavior of the driver, a human–vehicle–road model with steering torque as input is established. Then, command filtering adaptive control (CFAC) is used to design the steering control strategy of HMC, and Lyapunov theory is used to analyze the stability of the control system. To reduce the uncertainty range of the system and guarantee stability under disturbances, the unknown steering load boundary and driver's torque input are estimated. Finally, simulation and hardware-in-the-loop experiments verify the effectiveness of the proposed steering control strategy based on torque control. The experimental results show that the path tracking error converges to zero rapidly even when the HMC driving encounters disturbances. • The model we build describes driver behavior and vehicle dynamics. • Command filtering adaptive control is used to avoid the "explosion of complexity". • The unknown steering load boundary and driver's torque input are estimated. • System stability is ensured by estimating the boundary of unknown disturbances. [ABSTRACT FROM AUTHOR]