Total: 1
Biological specimens exhibit significant variations in size and shape, challenging autonomous robotic manipulation. We focus on the mouse skull window creation task to illustrate these challenges. The study introduces a microscopic stereo camera system (MSCS) enhanced by the linear model for depth perception. Alongside this, a precise registration scheme is developed for the partially exposed mouse cranial surface, employing a CNN-based constrained and colorized registration strategy. These methods are integrated with the MSCS for robotic micromanipulation tasks. The MSCS demonstrated a high precision of 0.10 mm $\pm$ 0.02 mm measured in a step height experiment and real-time performance of 30 FPS in 3D reconstruction. The registration scheme proved its precision, with a translational error of 1.13 mm $\pm$ 0.31 mm and a rotational error of 3.38$^{\circ}$ $\pm$ 0.89$^{\circ}$ tested on 105 continuous frames with an average speed of 1.60 FPS. This study presents the application of a MSCS and a novel registration scheme in enhancing the precision and accuracy of robotic micromanipulation in scientific and surgical settings. The innovations presented here offer automation methodology in handling the challenges of microscopic manipulation, paving the way for more accurate, efficient, and less invasive procedures in various fields of microsurgery and scientific research.