Autonomous Robotic Manipulation: Real-Time, Deep-Learning Approach for Grasping of Unknown Objects

Recent advancement in vision-based robotics and deep-learning techniques has enabled the use of intelligent systems a wider range applications requiring object manipulation. Finding robust solution for grasping autonomous manipulation became focus many engineers is still one most demanding problems modern robotics. This paper presents full pipeline proposing real-time data-driven approach robotic unknown objects using MATLAB convolutional neural networks. The proposed employs RGB-D image data acquired from an eye-in-hand camera centering interest field view visual servoing. Our aims at reducing propagation errors eliminating need complex hand tracking algorithm, segmentation, or 3D reconstruction. able to efficiently generate reliable multi-view grasps regardless geometric complexity physical properties question. system architecture enables simple effective path generation control. In addition, our modular, reliable, accurate both end effector We experimentally justify efficacy effectiveness overall on Barrett Whole Arm Manipulator.