Abstract: The intellectualization of coal mine is the core technology to realize the high-quality development of coal industry. The coal-gangue sorting robot can realize the automatic sorting of coal and gangue, which is an important equipment to realize the coal mine intelligent washing. On the basis of reviewing the research and practice results of coal gangue sorting robot and image recognition at home and abroad. In order to solve the low speed and low intelligence of coal gangue sorting robot, a delta robot with image recognition system was adopted to coal gangue automatic separate, which had been widely used in packaging and sorting industry. The work principle and design scheme of this robot were expounded in detail, the hardware devices such as industrial cameras, light sources, lens, etc were selected. A SVM recognition algorithm based on the fusion of grayscale and texture features was proposed. The software design method of robot control system was introduced. Moreover, an experimental platform was built for conducting online coal gangue sorting experiments. The results shown that the parallel robot could not only identify coal and gangue effectively, but also grasp the gangue less than 6 kg by controlling the pneumatic manipulator. Furthermore, the key technologies of the robot were put forward, including the rigid flexible coupling dynamic model and the rule of errors transfer, the coal gangue recognition algorithm based on the multi-feature fusion and deep learning, and the adaptive stability control strategy for parallel robot with external load disturbance. The research results in this paper will accelerate extensive application of the coal-gangue sorting parallel robot, and also provided a new way to realize the intellectualization and unmanned washing of coal mine.