Autonomous Robotic Bone Micro-Milling System with Automatic Calibration and 3D Surface Fitting

TL;DR

This work tackles the challenge of fully automating mouse cranial window bone milling by introducing a Microscopic Stereo Camera System (MSCS) and an online pre-measurement pipeline that performs automatic calibration and 3D surface fitting from uncalibrated views. The pipeline yields a rigid transform $oldsymbol T_{ ext{ ext{Mi}} ightarrow ext{R}}$ and a 3D‑fitted initial trajectory $oldsymbol p_{ini}$, enabling rapid, surface-adaptive milling when integrated with intraoperative feedback. Validation on euthanized mice demonstrates an 85.7% success rate with an average milling time of $2.1$ minutes, outperforming prior approaches in speed, accuracy, and stability. The results underscore the potential of end-to-end autonomous cranial window creation and offer a framework transferable to other microsurgical bone milling tasks requiring high precision and tissue preservation.

Abstract

Automating bone micro-milling using a robotic system presents challenges due to the uncertainties in both the external and internal features of bone tissue. For example, during mouse cranial window creation, a circular path with a radius of 2 to 4 mm needs to be milled on the mouse skull using a microdrill. The uneven surface and non-uniform thickness of the mouse skull make it difficult to fully automate this process, requiring the system to possess advanced perceptual and adaptive capabilities. In this study, we address this challenge by integrating a Microscopic Stereo Camera System (MSCS) into the robotic bone micro-milling system and proposing a novel online pre-measurement pipeline for the target surface. Starting from uncalibrated cameras, the pipeline enables automatic calibration and 3D surface fitting through a convolutional neural network (CNN)-based keypoint detection. Combined with the existing feedback-based system, we develop the world's first autonomous robotic bone micro-milling system capable of rapidly, in real-time perceiving and adapting to surface unevenness and non-uniform thickness, thereby enabling an end-to-end autonomous cranial window creation workflow without human assistance. Validation experiments on euthanized mice demonstrate that the improved system achieves a success rate of 85.7 % and an average milling time of 2.1 minutes, showing not only significant performance improvements over the previous system but also exceptional accuracy, speed, and stability compared to human operators.

Figures (8)

• Figure 1: The hardware setup of the autonomous robotic bone micro-milling system, including a robotic arm holding a microdrill, a head holder for the mouse to be milled, and a Microscopic Stereo Camera System (MSCS).
• Figure 2: End-to-end automation workflow of the improved autonomous robotic bone micro-milling system: the red part, Proposed Online Pre-measurement Pipeline, including Automatic Calibration (Section \ref{['sec:autocali']}) and 3D Surface Fitting (Section \ref{['sec:initraje']}), is the main contribution of this work and responsible for surface measurement before milling; the black part, Intraoperative Real-time Feedback, including Completion Level Recognition (Section \ref{['subsec:completionlevelrecognition']}) and Trajectory Planner (Section \ref{['subsec:trajectoryplanning']}), is improved from the original system zhao2023 and responsible for closed-loop control during milling. “Robot” refers to the robot motion controller, which controls the robot’s end effector (microdrill) to reach the specified coordinates. The robot coordinate system {R} and the microscopic 3D coordinate system {Mi} are illustrated at the bottom.
• Figure 3: The Automatic Calibration determines the rigid transformation matrix $\boldsymbol T_{\text{\{Mi\}}\rightarrow \text{\{R\}}}$ from the microscopic 3D coordinate system {Mi} to the robot coordinate system {R}. The robot executes a sequence of predefined movement commands to reach calibration setpoints, at each of which the 3D reconstructed drill tip point is obtained through 2D Keypoint Detection and 3D Reconstruction. Landmark Calibration then uses these 3D points and setpoints to calculate $\boldsymbol T_{\text{\{Mi\}}\rightarrow \text{\{R\}}}$.
• Figure 4: Schematic of 3D Surface Fitting. The bregma and lambda are detected through a CNN, and the milling center $\boldsymbol p_{c}^{\{\text{Mi}\}}$ is calculated by linearly weighting these two points. Subsequently, surface-fitted trajectory $\boldsymbol p_{fit}^{\{\text{Mi}\}}$ are sampled on the target surface based on $\boldsymbol p_{c}^{\{\text{Mi}\}}$ and radius $r$, and the 3D-fitted initial trajectory $\boldsymbol p_{ini}^{\{\text{Mi}\}}$ is obtained by offsetting $\boldsymbol p_{fit}^{\{\text{Mi}\}}$ along the normal vectors $\boldsymbol n$ by a distance $r_d+\Delta d$.
• Figure 5: Schemas of (a) the Trajectory Planner in the previous system, and (b) the Trajectory Planner in this study.