Robotic Bin-Picking System Based on Voice Recognition, Deep Learning, and Point Cloud Processing

This paper presents an automatic object localization system, which is used to pick the random and on-demand objects in the workspace. The system includes a robotic arm system integrated with a RealSense camera. Firstly, the target object is estimated from the speech recognition algorithm. Secondly, the Yolo-V3 algorithm is applied to detect and classify the target from the color image. Then, individual feature point clusters were extracted using segmented 2-D features and depth maps. To determine the position and orientation of the target, each cluster is matched to the CAD model using the ICP algorithm. Finally, collision avoidance techniques are applied to select objects for the picking task. The feasibility and effectiveness of the developed system have been verified experimentally. The test ended again showing that the system was able to successfully locate and pick up 3-D target objects via voice commands.