Robots are learning to perform surgery by watching videos of experienced surgeons. Two recent studies published in Science Robotics demonstrate autonomous systems completing complex procedures, from gallbladder removal to multiple laparoscopic tasks, without human intervention.
The Johns Hopkins Gallbladder Robot
A robot called SRT-H performed a key phase of gallbladder removal on eight unseen ex vivo pig gallbladders with a 100% success rate. The phase involved 17 tasks, focused on the clipping and cutting of the cystic duct and artery.
The robot learned through imitation. Researchers at Johns Hopkins, including Axel Krieger, showed the system demonstrations of gallbladder procedures on pig gallbladders. A high-level planner that works in language space issues the surgical steps as instructions, and a low-level imitation-learning policy turns those instructions into robot movements; the paper also tests GPT-4o in that planning role.
Performance was comparable to expert surgeons, though the robot took longer. It handled unexpected challenges, including anatomical variations and appearance changes during the operation.
Generalized Surgical Intelligence
A separate team from Chinese University of Hong Kong, Cornerstone Robotics, and Johns Hopkins tackled a harder problem: building a system that generalizes across different surgical tasks rather than mastering just one.
Their approach combines visual parsing using depth estimation and image segmentation, reinforcement learning, and a technique called zero-shot transfer that moves skills from simulation to real surgery.
Testing included seven game-based skill tasks on the da Vinci Research Kit, five surgical assistive tasks with the Sentire system on ex vivo animal tissue, and three tasks validated in live-animal trials. The system handled variations in scene layout, object sizes, instrument types, and lighting conditions.
What This Means
These are still research systems, not clinical tools. The Johns Hopkins robot worked on lifelike models, not human patients. The generalized system required live-animal trials to validate simulation-to-reality transfer.
But the trajectory matters. Previous autonomous surgery research focused on specific, constrained tasks. These two Science Robotics papers show robots picking up surgical skills, then adapting to novel situations.
The practical implications cut two ways. Autonomous surgical robots could extend expert-level surgery to places without enough surgeons, standardize procedures, and reduce human fatigue. They could also displace surgical work, raise liability questions, and introduce new failure modes.
The Fine Print
Neither system has regulatory approval for human surgery. The path from research demonstration to FDA clearance is long and uncertain.
Current surgical robots like Intuitive’s da Vinci system are teleoperated: surgeons control every movement through a console. These autonomous systems represent a different category that regulators have not yet addressed.
The Johns Hopkins research was funded by ARPA-H, the federal agency focused on health breakthroughs, along with NIH and NSF. The generalized surgical intelligence work involved collaboration between academic labs and Cornerstone Robotics, a company developing surgical systems.
Full autonomy for complex procedures remains further out, limited less by technical capability than by safety validation and regulatory frameworks.