Gaussian Process Regression for Sensorless Grip Force Estimation of Cable Driven Elongated Surgical Instruments.

Area of Science:

• Robotics
• Surgical Technology
• Machine Learning

Background:

• Haptic feedback is crucial for robotic minimally invasive surgery but is currently lacking.
• Accurate gripping force estimation is essential for safe and effective robotic surgery, especially with cable-driven instruments.

Purpose of the Study:

• To propose and evaluate Gaussian Process Regression (GPR) based methods for estimating gripping force in cable-driven surgical instruments.
• To compare the performance of different GPR configurations against traditional dynamic model-based approaches.

Main Methods:

• Developed four GPR-based filters: 2D GPR, 3D GPR, 2D GPR-Unscented Kalman Filter (UKF), and 3D GPR-UKF.
• Evaluated methods on a 10mm gripper using the Raven-II surgical robot platform.
• Compared GPR methods against a dynamic model-based UKF filter.

Main Results:

• All four proposed GPR methods outperformed the dynamic model-based UKF filter in precision and reliability.
• The simplest GPR filter with 2-dimensional inputs demonstrated the best performance among the evaluated methods.
• GPR methods achieved superior results without requiring parameter tuning.

Conclusions:

• GPR-based gripping force estimation is a viable and effective solution for robotic minimally invasive surgery.
• The 2D GPR filter offers a promising, high-performance approach for enhancing haptic feedback in surgical robotics.
• The proposed GPR schemes improve the precision and reliability of surgical robotic systems.