Abstract: Aiming at the current mainstream coal and gangue sorting method, there was problems such as large waste of resources, heavy environmental pollution and low sorting efficiency. A multi - manipulator collaboration coal and gangue sorting robot was proposed, and the multi-dynamic target and multi-manipulator cooperation method of the robot was studied. In order to solve the problem of gangue identification in industrial environment, a rapid identification method based on deep learning network was proposed, which realized real-time identification of coal gangue flow on picking tape machine, and effectively improves its comprehensive accuracy. The relative coordinates and depth information of gangue were obtained in real time by binocular vision technology, And the corresponding error analysis and error compensation methods of 3D information were studied, which provide a basis for the coal and gangue sorting of the robot. In order to achieve efficient sorting of multiple gangue in the sorting area, a multi-dynamic target,multi- manipulator coordinated coal,gangue sorting strategy and corresponding sorting process were proposed. The self-planning test results of the manipulator picking trajectory were realized. The rapid identification and positioning method of the target recognition accuracy rate under the mixed sample reached 76.92%. The multi-arm collaborative sorting method significantly improved the efficiency of coal gangue sorting compared with the single-arm method. In summary, the coal-gangue sorting robot based on multi- manipulator collaboration provided a feasible method for realizing the automatic and intelligent sorting of coal and gangue, reducing the investment and production cost of coal washing and improving coal quality.