Abstract: In view of the current research on robot welding of workpieces with uncertain geometric and positional information based on 3D visual sensing, the method for determining the spatial posture of the welding torch at the welding path point is highly targeted and has high limitations. A method of determining welding torch space pose based on local point cloud of weld path points was proposed. Firstly, the spatial pose parameter model of welding torch is described, and then the local point cloud at the weld path point is obtained by considering the the sensor front-mounted distance to determine the minimum search radius of kd-tree, and the deflection plane and inclined plane of welding torch at the path point are gradually determined, so as to determine the non-collision space pose between the welding torch and the workpiece at the weld path points. The results show that this method has strong adaptability and good effectiveness. Due to the surface quality of point clouds, there is certain fluctuation in the attitude determination results compared with the real attitude, but it can be smoothed to meet the needs of robotic welding. This method is expected to improve the level of robotic autonomous welding operation based on 3D visual sensing for workpieces with uncertain geometric and positional information.