SimRetract: trained only in simulation, a robotic arm resumes the surgeon's hand spreader in open surgery

Without any human demonstration or field adjustment, the robot captures the spreader in 90% of the phantom cases and maintains the gap when passing in 78%.

In open surgery, holding a retractor to maintain the operating field unobstructed requires a hand or an assistant for long minutes. One team presents what it describes as the first direct transfer from simulation to reality for this surgeon-robot collaboration: the arm takes the retractor from the surgeon's hands and maintains the distance at the time of handover.

The control policy has only seen simulated demonstrations, designed to transfer well: the robot observes images where the instrument is masked and its grip geometry, learns to catch itself from unusual positions and keeps a margin against execution drifts. Out of 50 presentations defined by phantom surgeons, it grips the instrument in 90% of cases and preserves the gap in 78%. Tests on unknown retractors give hope of a generalization to instruments with a similar grip.

Out of ten configurations and 1,200 trials, the authors isolate what makes the gap between simulation and reality: removing the instrument mask alone causes the real robot to drop to 22% of grips and 12% of holdings, while all variants remained between 80 and 96% in simulation.