Vision-based tile recognition algorithms for robot grasping task in EAST

摘要

During the operation of the experimental advanced superconducting tokamak (EAST), files installed on the vacuum vessel break and fall onto the divertor. To ensure high experimental performance, a robot removes the broken files and replaces them with fine files. However, one of the key issues for this task is to distinguish fallen files from other files for grasping. In addition, discrete spots caused by burn and corrosion on the files' surface also bring challenges in vision-based recognition. Here, a novel breakpoint detection algorithm is proposed that recognizes fallen files. Experiments are performed to demonstrate the recognition and catch processes of fallen files and to test the robustness of the algorithm.