3D camera equipped robotic arm for small object manipulation

Abstract:

In recent years, due to the advancement of computer vision and artificial intelligence, robotic arm systems are commonly equipped with a camera for target detection and localization. However, locating and grabbing small objects is still challenging due to low resolution and image noise. To mitigate this issue, we propose to build a 3D stereo camera equipped robotic arm system which consists of Intel RealSense D405, NVIDIA Jetson Board, and Mirobot’a desktop robotic arm. RealSense D405 provides sub-millimeter accuracy for close-range: it operates at a range of 7 to 50cm with minimum object detection down to 0.1mm at 7cm. Utilizing the depth cues to assist small object segmentation is investigated in this study. Moreover, depth-enhanced deep learning algorithms are implemented on NVIDIA Jetson Board and its performance is analyzed in terms of speed and accuracy. Lastly, Mirobot is controlled by the NVIDIA Jetson Board via USB and tested for small object manipulation in the experiments.