Abstract: In the grinding force control process for complex surfaces, the fluctuation of grinding force caused by the uncertainty of workpiece curvature will greatly affect the surface quality of parts. In order to solve this problem, a robot force control method is proposed by combining variable integral PID (proportional-integral-differential) feedforward controller and adaptive control algorithm. According to the contact force signal collected by the 6-dimensional force sensor and the robot end position signal sent by the upper computer, the robot motion bias is modified online to implement accurate force control on the robot end, thereby reducing the force tracking error caused by the uncertainty of operating information. The comparison between simulation and experimental data shows that the proposed method can quickly respond to the change of workpiece parameters, effectively reduce the tracking error of contact force, and shorten the steady-state adjustment time, which makes up for the defect that the traditional position-based impedance control strategy cannot obtain workpiece parameters in advance. Experimental results show that the force control method proposed can effectively improve the force tracking performance of position-based impedance control with constant parameters, so that the grinding roughness of the surface with variable curvature is reduced from about 0.6 μm to nearly 0.4 μm, and the robot grinding performance is improved.