Autonomy, Operating Domains, and Validation in Robotic Ultrasound: Transferable Lessons and Structural Limits from Autonomous Vehicle Development

Abstract

Diagnostic ultrasound is among the most portable and least expensive of the major imaging modalities, yet its reach is constrained by the availability and physical capacity of skilled operators: between 85 and 90 percent of sonographers report work-related musculoskeletal pain, and approximately one in five sustains a career-ending injury [1, 2, 3]. Robotic ultrasound systems have advanced substantially in response and learned acquisition policies now approach expert-level performance on individual anatomical targets [4, 5]. Progress in the field is, however, characterized using autonomy taxonomies inherited from surgical robotics [6, 7, 8]. These are frameworks that describe how much of a task a machine performs while remaining largely silent on the population in which that performance was demonstrated. Across the studies examined in this review, performance metrics are routinely reported, whereas explicit specification of operating conditions is considerably less common.