Abstract: Aiming at the problems of cleaning robots on the structured road surface in the parking lot, such as excessive acceleration and large path deviation from the global path, an improved dynamic window approach (DWA) is proposed. In order to limit the acceleration range of the vehicle, the dynamic constraints of the DWA speed space are optimized firstly to avoid the situation that the excessive acceleration leads to very small vertical tire load. Then, error compensation for trajectory estimation is performed in real time based on the laser odometer, to solve the problem that the path deviates greatly from the global path on the parking lot road. Finally, the improved DWA is applied to a cleaning robot with four wheels of independent driving and independent steering to conduct comparative experiments. Experimental results show that the average path error of the improved DWA is reduced by about 60% compared with the traditional DWA on the basis of the same global path and road conditions, and the vertical load of tires is also greatly increased under different road conditions, which verify the effectiveness and reliability of the proposed method.