Abstract: Highly nonlinear units such as deadzone and backlash exist in the joints of mobile repair welding robot, which have negative effects on control accuracy. To improve the path control accuracy of the end effector, Particle swarm optimization algorithm is used to identify the model of the nonlinear joints system. Then switching compensation control method referring the identified model is applied to the joint controlling system, combined with feedfoward-feedback control based on inverse kinematics, the hybrid control method realize considerable path accuracy. Experiments show that the average path error of designed welding robot's end effector is less than 0.2 mm with this control method, and the shortcomings of low precision reducer are compensated.