Abstract: In order to improve the robot’s accuracy when cutting workpieces, a path planning study was carried out on a 3D vision cutting robot. A multi-point calibration method that solves the hand-eye relationship matrix was used to calibrate the camera according to the fixed view field of the 3D camera. The point cloud data obtained by the 3D camera sampling was transformed into point cloud data in the robot coordinate system. A threshold adaptive sampling method was proposed for point cloud data processing, and the pre-cut spatial curve was fitted using cubic polynomials. An attitude estimation algorithm was used to complete the calculation of the forward vector and feed vector in the cutting robot’s operating space by fitting the cutting microtangent plane, and the planning results were applied to the robot’s cutting operation. The experimental results show that the error of the system is within 0.24 mm and the cutting trajectory is well aligned, which can meet the requirements of the cutting robot system.