This paper designs a robot autonomous sorting system for intelligent logistics. Firstly, the current frame image is captured by the camera, the package capture image features and letter information are obtained by SIFT (Scale Invariant Feature Transform) algorithm, and the coordinates of the image features are transformed into the robot workpiece coordinate system, so as to obtain the optimal grasping attitude of the manipulator. At the same time, the manipulator grabs and places the target package on the mobile robot; Secondly, multiple mobile robots use path planning combined with scene 2D grid map to plan a flexible and autonomous optimal collision free path, so as to transport each package to the place where its letters represent, and finally complete the unmanned sorting of the warehouse. The experimental results show that the robot autonomous sorting system designed in this paper can accurately grasp the target package, quickly avoid static and dynamic obstacles and reach the target position independently, which can further improve the intelligent level and distribution efficiency of unmanned warehouse.