In context
In 2024, retinal surgery faced a global shortage of skilled surgeons, compounded by an aging population and rising blindness rates. While robotic assistants improved precision by filtering hand tremor, they still required expert control. This paper addressed the next step: enabling robots to learn and autonomously reproduce expert surgical motions.
What was reported
Researchers at Beihang University proposed a method to acquire and learn a doctor's fundus injection skills. They built a motion-capture platform using two Kinect depth cameras and ArUco markers on the surgeon's hands to record 3D trajectories during simulated procedures. A multi-camera setup with an ArUco box enabled coordinate fusion to correct occlusions and maintain accuracy.
To segment the continuous motion into meaningful phases, they applied a Hidden Markov Model (HMM) with five hidden states representing surgical intentions. Using Baum-Welch for parameter estimation and Viterbi for state decoding, the trajectory was divided into approach, insertion, injection, withdrawal, and return phases.
For skill learning and reproduction, they employed imitation learning and validated the approach on a RealMan RM65-B robot. Results showed the robot could effectively represent and learn the surgical skills, reproducing the demonstrated operations.
Why it mattered
This work moved beyond teleoperated assistance toward autonomous execution of delicate retinal procedures. By encoding expert skills into a robot, it could reduce the dependency on scarce specialists, shorten learning curves, and potentially improve consistency in microsurgery—an important step for intelligent surgical robotics in clinical practice.
“The proposed imitation learning method for fundus surgical behavior can effectively realize the representation and learning of surgical operation skills, and guide the robot to complete the relevant surgical operations.”
Source: 《机器人》期刊 (robot.sia.cn) · Published 2024-05-08 · “基于模仿学习的眼底手术行为机器人复现”