Robot Imitation Learning for X-ray-Guided Spine Surgery Explored

A recent study explores the potential of using imitation learning for robot control in autonomous X-ray-guided spine surgeries, focusing on cannula insertion during vertebroplasty. The researchers created a sophisticated in silico environment for the scalable and automated simulation of bi-plane-guided spine operations with high realism. They compiled a dataset featuring accurate trajectories alongside corresponding bi-planar X-ray sequences that mimic the stepwise alignment performed by practitioners. The imitation learning policies were developed to facilitate planning and open-loop control, successfully aligning a cannula based solely on visual data on the first attempt, shedding light on the strengths and limitations of this approach for procedures reliant on sparse X-ray input.