Speaker
Description
Handling radioactive isotopes in the CSNS Phase-II project requires high-precision automation to minimize human radiation exposure. The final deployment will feature a heavy-duty robotic arm mounted on a ground track operating in highly constrained, shielded environments. However, as the final hardware is not yet deployed, this abstract presents a lab-scale validation framework using an AUBO 6-DOF collaborative robotic arm and a RealSense depth camera to de-risk software and algorithm development.
Our core work focuses on vision-based pose estimation and Eye-in-Hand visual servoing. To extract stable 3D coordinates from depth images, we implemented multi-frame filtering and fixed working-plane constraints. Furthermore, to address the limited field of view in narrow storage spaces, we developed an iterative, PID-like visual servoing control logic. The robotic arm actively adjusts its pose to “pull” the target back to the image center, ensuring robust target tracking and alignment during the approach phase.
By decoupling perception and execution, the validated algorithms—including object detection, coordinate transformation, and visual servo logic—are hardware-agnostic and directly transferable to the final track-mounted system. This lab-scale methodology significantly reduces deployment risks and provides a scalable foundation for future integration into distributed control architectures.
