High-precision control of robotic arms based on active visual under unstructured scenes.

Eye-hand-robotic arm has the characteristics of dexterous manipulation, flexible movement and high precision under structured scenarios. However, it is still a difficult problem for robotic arms to achieve high-precision manipulation in complex unstructured environments. In this paper, we propose an active vision-based motion control approach that enables the robotic arm to realize the high-precision manipulation. We divide the motion of the hand-eye arm into two stages: in the first stage, the robotic arm exploits a position-based visual servo strategy to realize flexible motion with 6-DOF by fast-tracking the marked target within its moving range; In the second stage, the micro-target is recognized based on convolution neural networks, and then the visual pixel errors are used to map the joint space of the robotic arm, and to further complete the operation task. Experiments results show that the control accuracy of the manipulator can reach the millimeter level by using the method proposed in this paper.