Humanoid robots show early promise in clinical procedures, but technical limits remain

Editor's Note

Adapting general-purpose humanoid robots already trained on large-scale industrial tasks for use in surgery could provide a solution to burnout, understaffing and other problems. This is the primary takeaway of a July 9 article from UC San Diego detailing the work of robotics expert Michael Yip, who has already published a perspective piece in Science: Robotics. 

As detailed in the article, equipping surgical robots with arms and multi-fingered hands would enable them to take advantage of powerful AI foundation models used in other industries. This would allow for greater adaptability and faster skill acquisition in clinical environments. Humanoid robots could assist with common but physically demanding tasks, such as holding ultrasound probes or endoscopic cameras during procedures or passing instruments as a scrub nurse.  

The research is outlined in more detail in the upcoming paper “Humanoids in Hospitals: A Technical Study of Humanoid Surrogates for Dexterous Medical Interventions.” Researchers developed a teleoperation system for the Unitree G1 humanoid robot and tested it across seven hospital tasks, including auscultation, bag-valve mask (BVM) ventilation, intubation, tracheostomy, ultrasound-guided injection, and suturing. The system integrated dual-arm control, high-fidelity hand tracking, force-modulating impedance control, and customized grasp configurations. Non-clinician operators guided the robot through various procedures using VR trackers and a multi-camera interface.

The full article offers extensive detail on how the robot performed. Overall, while robotic consistency in tasks like BVM squeezing is promising, clinical application reportedly remains limited by force generation, sensor sensitivity, and dexterity. The researchers advocate further development in robotic feedback systems and more realistic training environments, as well as formal trials to validate safety and effectiveness in real-world settings.