Experimental AI endoscope navigates gastrointestinal constrictions in phantom trials

Researchers introduced StenoVLA-3D, an experimental system that uses three-dimensional geometry and prior observations to guide an endoscope through constricted gastrointestinal passages. In 36 physical trials for each organ type, it completed 88.9% of esophageal tasks and 77.8% of colonic tasks. The system builds spatial point maps from camera observations, tracks traversal progress and generates a lesion report after a lesion leaves view. Its evidence comes from recorded episodes and a three-degree-of-freedom endoscope operating in phantoms, not patients.