CLOVER Lab

The identification and grasping of random and disordered objects by robotic systems is a challenge. With the update of scenarios, objects tend to appear in a random state, which makes the classification of vision and the operation of robots difficult. We simulate the process of object generation in the physics simulator and randomize it with real-world object information. By means of virtual-reality fused, a large amount of data are generated, which is used for training the neural network to recognize disordered objects. Then, based on the accurate classification and the position pose information, the robot will grasp through the optimized motion algorithm. Meanwhile, we also design end-effectors for random disordered objects, which are highly adaptable to objects with discrepant shapes, types, and materials.