In order to overcome the limitation of the traditional terahertz tomography system which can only be applied to flat samples, a robot-based terahertz tomography system is built by combining an industrial robot and a reflective terahertz time-domain spectroscopy system. Using the robot graphical offline programming technology to plan the scanning path, it can be ensured that the terahertz wave is perpendicular incident on the sample surface and the maximum reflective signal can be received. The deconvolution algorithm is used to process the terahertz time domain signal to improve the time resolution capability of the system. The rubber material with the shape of a curved surface is selected as the sample, and the normal vector of each sampling point on the sample surface is calculated by the principal component analysis method, and the three-dimensional reconstruction of the sample surface and internal structure is carried out with the time-of-flight imaging technology, which verifies the feasibility of the system.