Abstract: In order to make the probe end of the robotic arm always be perpendicular to the surface of the tested workpiece during the scanning process, an adaptive detection algorithm for curved surfaces based on normal feature extraction was proposed. The 3-D data of the surface of the test piece was acquired by the optical scanning device. The mapping function used for pose correction was derived by the normal extraction algorithm. Finally, the adaptive planning of the scan path was completed. In MATLAB, stray points were removed from the sample point cloud, and the three-dimensional reconstruction of the test piece was completed through the QUALIFY software. In the experiment, a curved sample block with a size of 150mm×200mm×10mm was optically scanned and path optimized. The test results show that the maximum deviation of the position test accuracy in the x-axis, y-axis, and z-axis directions are 0.5142mm, 0.2645mm, and 1.4265mm respectively. The maximum value of the overall position deviation is 1.1135mm, and the average value is 0.5647mm. After substituting the deviation as a compensation parameter into the planned path of the manipulator, the system realizes the online adjustment of the probe pose during the scanning process. This research is of great significance to the application fields that require the probe to be strictly perpendicular to the tested workpiece in the automated test system.