UCSD's humanoid surgery test puts cheap robots in da Vinci's lane

Why it matters

UC San Diego's study turns humanoid surgery from a speculative pitch into a preclinical benchmark, while showing how far cheap general-purpose robots remain from hospital use.

Michael Yip and a UC San Diego team reported the first live-animal surgery performed with humanoid robots, using two teleoperated Unitree G1 machines to complete laparoscopic gallbladder removal on a pig.

The study, published in Nature in July 2026 and covered by Forbes on July 10th, is best read with its constraints in the first paragraph. The patient was a pig. The robots were controlled by trained human surgeons. The machines were tethered so they could not fall onto the operating table. This was preclinical research, not a clinical deployment.

Those caveats make the result more useful, not less. Yip, who leads UC San Diego's Advanced Robotics and Controls Laboratory (ARClab), has worked on surgical robots, robot learning and surgical automation. The July paper pushes that work into a sharper commercial frame: whether general-purpose humanoids can take on tasks that have belonged to expensive, purpose-built surgical systems.

In the live-animal portion, the robots handled the sequence that matters in a laparoscopic cholecystectomy: tissue retraction, dissection, clipping and gallbladder removal from the liver bed.

The cheap robot is the provocation

The researchers did not build a bespoke surgical machine. They used off-the-shelf Unitree G1 humanoids, according to the project materials and Forbes' account. Unitree's own shop lists the G1 at $13,500, before shipping, customs and any configuration changes. Forbes described the robots used in the study as customized G1s, roughly four to five feet tall and about 70 pounds, and said they likely used Unitree's dexterous hand setup.

That price point is the reason this experiment has startup relevance. Intuitive Surgical's da Vinci business has been built around specialized systems, proprietary instrumentation, training, service and hospital workflows. That installed base is a moat made of regulatory history, surgeon training, procedure volume and hospital purchasing behavior.

A $13,500 humanoid cannot walk into that market because it performed one animal procedure. The G1 test asks a different question: how much of surgical robotics must remain purpose-built, and how much can shift to cheaper general-purpose hardware paired with better teleoperation, perception and controls?

Yip's lab has been asking that question for years in adjacent forms. ARClab's surgical robotics page describes work on new surgical robots, real-time medical image guidance and task automation. Its opportunities page specifically calls out humanoid robots among the platforms used for surgical automation research. The July paper gives that agenda a tangible artifact: two small humanoids on a live porcine surgical case, doing work that normally implies a much heavier capital stack.

Teleoperation keeps the human in control

The Nature abstract is explicit that the study is an evaluation of humanoid technology for laparoscopic surgical tasks. It does not claim autonomous surgery. It says the team developed a humanoid-based laparoscopic teleoperation framework using general-purpose instruments, then compared the results against clinical-readiness requirements and established surgical platforms.

That distinction matters because the commercial promise of humanoid robots often gets compressed into autonomy. In surgery, the first useful version may be remote physical presence. A specialist could operate a teleoperated system in a hospital that lacks the same surgical staffing. A humanoid form factor could, in theory, use existing instruments and hospital layouts instead of forcing every care site to buy and train around a fixed surgical station.

The Nature paper frames the motivation around healthcare staffing shortages and rising care demand. The authors argue that much hospital work is physical, requiring mobility, manipulation and safe interaction in spaces designed for people. A humanoid that can move around those spaces and hold tools could eventually be useful beyond the operating room. Surgery is the hardest version of the argument, because the tolerance for latency, drift, imprecision and failure is close to zero.

The paper also states that key technical challenges remain before clinical deployment. That sentence does a lot of work. A live-animal surgery proves feasibility in a constrained setting. It does not answer sterilization, reliability, maintenance, credentialing, liability, reimbursement, surgeon training or regulatory review. It also does not answer whether a commodity humanoid can be hardened for a hospital network.

That last point is not theoretical. The National Vulnerability Database lists CVE-2025-35027, a 2025 command-injection vulnerability affecting the Unitree G1. That does not mean the robots used in UC San Diego's experiment were compromised or unsafe. It means any path from lab demo to medical device would require a security posture very different from ordinary research hardware.

The incumbents are already moving

The timing lands during a busy period for surgical robotics. Medtronic said on December 3rd, 2025, that the FDA cleared its Hugo robotic-assisted surgery system for U.S. urologic procedures, including prostatectomy, nephrectomy and cystectomy. Johnson & Johnson said on January 7th, 2026, that it submitted the OTTAVA robotic surgical system to the FDA for De Novo classification in multiple upper-abdominal general surgery procedures.

SS Innovations has been pushing the remote-care angle even harder. In April, the company described the SSi Avtara Humanoid Surgical Platform as a conceptual humanoid system for surgical applications across healthcare, defense, disaster response and industrial settings. It also described mobile and remote surgical systems designed for settings far from standard operating rooms.

UC San Diego's work sits apart from those company road maps because it uses a general-purpose humanoid already available on the market. That makes the result less polished and more interesting. The researchers are testing whether the commodity robotics curve, the same curve driving warehouse, home and industrial humanoid efforts, can bleed into a medical device category that has long moved at the pace of clinical validation.

Humanoid builders increasingly bet on one body plan accumulating enough experience to work across many human spaces. UC San Diego has now tested the most demanding version of that thesis under surgical conditions, with trained physicians still doing the operating.

What the test actually changes

The study gives humanoid robotics founders a better target than demo reels. A humanoid that can dance, carry boxes or fold laundry proves motor control in forgiving environments. A humanoid that can hold laparoscopic tools, maintain a surgical workspace and follow a human surgeon through a gallbladder removal on living tissue faces a much harsher set of constraints.

It also gives surgical robotics companies a reason to watch the low end. Intuitive, Medtronic and J&J are not threatened by a Unitree G1 in a university lab. Their businesses are protected by clinical evidence, installed systems, trained surgeons, service networks and regulatory status. The pressure comes later, if cheaper humanoid hardware keeps improving while teleoperation software and surgical perception improve around it.

That is the bet in Yip's work: keep the surgeon in the loop, use humanoid hardware to reduce the cost and rigidity of patient-side robotics, and test the system against real surgical tasks rather than toy manipulation. The July paper does not make the robot surgeon autonomous. It makes the operating room a credible proving ground for general-purpose humanoids.

For founders, that is the commercial opening. The first company to benefit may not be the company that builds a full humanoid surgeon. It may be the one that solves a narrower bottleneck: safe teleoperation, sterilizable end effectors, surgical tool tracking, security, simulation, training, or workflow software for remote robotic procedures. UC San Diego's experiment shows the hardware can enter the room. The business will belong to whoever can make hospitals trust what happens after that.