Surgical robots are not new: the da Vinci system has been in operating theatres for more than two decades, and several private hospitals in Dhaka now advertise robot-assisted procedures. But a da Vinci does exactly what the surgeon’s hands tell it to do through the console. What a team at Johns Hopkins University has demonstrated is different in kind — a robot that watched surgeons work and learned to do the tasks itself.
Trained like a chatbot, on video instead of text
The method is called imitation learning. Instead of programming every movement step by step, the researchers fed the model recordings from the small wrist-mounted cameras on da Vinci robots, capturing thousands of examples of experienced surgeons manipulating needles, lifting tissue and suturing. The model learned the mapping from what the camera sees to what the instruments should do — the same basic recipe as a large language model, with video in place of words. Presented at the Conference on Robot Learning in Munich, the result is a robot that performs those three tasks autonomously, and, the team says, at the skill level of a human surgeon.
“The model is so good at learning things we haven’t taught it,” said Axel Krieger, the assistant professor of mechanical engineering who led the work. “Like if it drops the needle, it will automatically pick it up and continue.” That recovery is the striking part. Traditional surgical automation fails at the first unexpected event; a model trained on real, messy footage has seen surgeons drop things and carry on.
Talking to the robot
Team member Ji Woong “Brian” Kim added that the robot is built to take spoken direction — “move left”, “move right”, “do this” — much as a surgeon instructs a junior. That points to the realistic near-term use: not a robot operating alone, but one that handles the tedious, repetitive parts of a procedure while the surgeon supervises and steps in.
The limits, and the Bangladesh angle
The researchers themselves stress that this is not risk-free. A small error in suturing is a serious injury, and the model has been validated on training tasks, not on a full operation on a live patient. Regulators will want years of evidence before any of it runs unsupervised. For Bangladesh, where the shortage is not robots but trained surgeons — particularly outside Dhaka and Chattogram — the promise is longer-term and real: a system that can perform standard steps to a consistent standard could let one specialist supervise more procedures, and eventually more locations, than the two hands they were born with allow.
Source: Indian Express, via Prothom Alo.
