A cheap stock humanoid just did surgery. Now imagine what a high-end robot could do

I kinda think the most important number in the humanoid-surgery story that broke this week isn't a precision score or a success rate. 

It's the price tag: roughly $16,000.

(Maybe more, depending on the exact hands used.)

In a paper published July 8 in Nature, a team from the University of California, San Diego showed that off-the-shelf humanoid robots can perform the core steps of real surgery. Working under full teleoperation, the humanoids carried out the key stages of laparoscopic cholecystectomies — gallbladder removals — on live pigs: retracting tissue, dissecting, clipping the cystic duct and artery, and lifting the gallbladder out of the liver bed.

For a robotics audience, the setup is the interesting part. 

The team didn't commission a bespoke surgical machine like a da Vinci surgical robot, which can run millions of dollars. They grabbed Unitree G1s — one of the most basic and affordable humanoids you can buy — then put standard human surgical instruments in their hands, and wrapped the whole thing in a teleoperation framework they call LapSurgie. 

And it worked well enough to publish in Nature. It wasn't perfect, and it won't show up tomorrow at a hospital near you, but it showed a ton of promise.

The robot is almost aggressively unremarkable

The G1 is small and cheap by humanoid standards: about four and a half feet tall, roughly 75 pounds, and available for well under $20,000. It's the robot you see doing kung-fu demos and dance routines on YouTube. 

The specific units here were almost certainly running Unitree's three-fingered Dex3 dexterous hands — thumb, index and middle finger, each with multiple actuated joints — plus the extra wrist and waist articulation that makes holding and using tools possible. It also had standard onboard sensing (3D LiDAR, depth camera), though the actual view into the abdomen came from a normal laparoscope, not the robot's own eyes.

To state the obvious: this is not a surgical robot. 

It's a general-purpose humanoid that was teleoperated. 

And that's exactly why it matters.

Contrast it with the alternative. The benchmark in this field is Intuitive Surgical's da Vinci, the platform that has defined robotic surgery for two decades and sits in thousands of operating rooms. But da Vinci is a fixed, purpose-built system: wheeled in, docked, brilliant at one job, and priced on the order of a couple million dollars. 

It does not walk down a hall, and it's out of reach for a lot of small city and county hospitals.

It's not about whether a humanoid robot can beat da Vinci. Let's be honest: it can't, and probably won't for a long time. It is about "how far can the cheapest, most general machine already get?" The answer is farther than I would have thought possible.

Yes, there are caveats

This is a feasibility study, and the researchers frame it as one. The patients were pigs, not people. The robots were fully teleoperated — a human surgeon was driving the entire time, so nobody is claiming autonomy. 

The paper's own captions describe the work as "humanoid-assisted," so this isn't a claim that a robot ran a case solo from first incision to final closure. And the hard problems of surgical robotics are all still here, made harder by using a non-surgical body: holding a stable pivot at the incision (the remote center of motion), fighting tremor and latency, and operating inside the tight precision budget of working next to a liver.

None of that erases the milestone.

All of it does set the baseline.

Now imagine a top-end robot

If a bottom-of-the-range G1 can do this under teleoperation, the interesting extrapolation is ... what happens as the hardware improves?

Any of the serious humanoid makers today, including Unitree, has robots and hands better than the surgical test team. More dexterous, higher-DOF hands are already shipping. 

Give a teleoperated surgeon that kind of dexterity and the fiddly steps like fine dissection, suturing, and knot-tying get dramatically more doable. Add force feedback and haptics and the operator starts to feel tissue again, closing one of teleoperation's biggest gaps.

The cool thing here is the potential for a general-purpose robot that can move through a hospital built for humans, pick up tools designed for human hands, and take on many embodied tasks instead of one ... including being driven by an expert who is nowhere near the room. 

Roughly half the world lives somewhere with too few surgeons, and "medical desert" isn't only a developing-world problem: rural regions across the U.S. and Canada routinely send patients hours away for procedures a robot plus a remote specialist just might be able handle locally in a few years. 

Push it to the edge cases — a winter-over crew in Antarctica or an astronaut on the ISS — and a cheap, teleoperable humanoid stops looking like a hospital gadget and starts looking like essential, even core infrastructure for the places medicine can't currently reach.

Of course, that's still just the teleoperated version. 

The further-out bet is autonomy: a humanoid that downloads a surgical physical AI foundation model the way today's robots download locomotion policies and executes routine procedures with (maybe) a human supervising rather than driving.

How far into the future is that? We can almost start to see it now.