Abstract
This research presents an autonomous risk detection system designed to enhance the safety of robotic laparoscopic surgery. By utilizing force sensors located on the slave manipulator's motor side and employing machine learning, the system estimates the mechanical properties of manipulated tissues in real-time. Experimental results using a membrane-like proxy demonstrate that the system can successfully detect and halt imminent perforation risks, providing a critical safety layer that functions independently of communication stability between the surgeon and the robot.
Introduction
Modern robotic surgery typically operates in a master-slave configuration, where a surgeon's inputs at a console are transmitted to robotic manipulators inside the patient. While effective, this architecture is vulnerable to communication delays or errors. If the link is compromised, the "slave" robot may execute unintended motions, leading to accidental organ damage. To mitigate this, we propose a "smart slave" system capable of autonomous risk detection based on the physical interaction between the surgical tool and biological tissue.
Methodology
The proposed system identifies perforation risk through a three-stage computational pipeline:
- Signal Denoising: Because the force sensors are integrated into the motor unit rather than the surgical tip (to maintain sterilizability), the raw data contains significant mechanical noise and friction. A Kalman filter is applied to estimate the actual contact force at the instrument tip.
- Physical Property Estimation: The system models tissue as a linear spring-damper system. Using Recursive Least Squares (RLS), it calculates the elastic coefficient ($k$) and damping coefficient ($c$) in real-time.
- Risk Classification: A Support Vector Machine (SVM) classifier analyzes a feature vector comprising force, velocity, acceleration, and the estimated $k$ and $c$ values. It categorizes the interaction into three levels: Level 0 (Safe), Level 1 (Potential Risk), and Level 2 (High Risk/Imminent Perforation).
Experimental Results
To validate the concept, the system was tested using an 8μm thick polyvinyl chloride (PVC) film as a proxy for delicate biological membranes. The robot was programmed to perform axial motions against the film to gather training data and evaluate real-time avoidance.
The system demonstrated exceptional performance in real-time risk detection. In 10 out of 10 trials, the autonomous safety mechanism successfully halted the robot's motion before the film was punctured. Quantitative analysis showed a classification accuracy of 99.9% for safe states and 100% for high-risk states. A key finding was that the estimated elastic coefficient ($k$) rises sharply just before perforation, serving as a more reliable indicator than raw force data alone.
Conclusion and Future Work
This study confirms that autonomous, sensor-based risk detection can effectively prevent surgical accidents even when human-operator communication is unreliable. While the current model uses a simplified 1-DOF setup and linear tissue models, future work will focus on multi-degree-of-freedom movements and the complex, non-linear properties of actual biological organs. Moving toward unsupervised learning will also allow the system to adapt to different tissue types without extensive pre-training.
Summary updated