Endovascular interventions depend on physicians reading planar DSA and X-ray fluoroscopy to understand vessel anatomy and where the device is moving. That human interpretation step is hard to automate: the images are 2D projections, and converting them into feedback signals that an autonomous system can act on has remained an open problem.
A new preprint on arXiv (2610.21481) takes on that problem directly. The title describes a "2D image feedback-based autonomous endovascular navigation method," and the abstract frames it as a perception and control framework. In other words, the work appears to couple image understanding with the control logic needed to steer a device through the vasculature.
The abstract excerpt available is brief and cuts off mid-sentence, so the specific architecture, experiments, and results are not described here. What is clear is the motivation: removing the manual step of converting fluoroscopic images into navigation commands.