A humanoid robot performs a surgical operation for the first time in history

A humanoid robot performs a surgical operation for the first time in history

🔎 Why surgery has just entered a new era

On July 8, 2026, a study published in Nature documented an event the medical community had been waiting for for decades. Two teleoperated humanoid robots performed two gallbladder removals on live pigs during a preclinical trial conducted at the University of California, San Diego (UCSD).

This is not a specialized surgical robot like the Da Vinci. These are versatile humanoids — Unitree G1s, nicknamed "Surgie" by the team — measuring about 1.50 meters, initially designed for logistics or research. Their precision was sufficient to perform minimally invasive laparoscopic surgical procedures. The boundary between general-purpose robotics and specialized surgery has just collapsed.

The magnitude of this shift is hard to overstate. Until now, every advance in robotic surgery relied on single-function machines, confined to an operating room. Here, it is demonstrated that a generic humanoid can be piloted by a surgeon to complete an entire procedure. The implications for telesurgery, healthcare access, and remote deployment are considerable.

The key points

• Two teleoperated Unitree G1 humanoid robots performed two laparoscopic cholecystectomies on live pigs, a world first.
• The study is published in Nature on July 8, 2026, by a team from UC San Diego.
• A surgeon controlled the robots from a console, without direct intervention on the patient.
• This is a preclinical trial: no human was operated on. Clinical trials are yet to come.
• The robot used is not a dedicated surgical system but an adapted general-purpose humanoid.

Recommended tools

What actually happened at UC San Diego

Two complete surgeries, two successes. This is the blunt summary of the trial conducted by the UC San Diego team, reported by Ars Technica.

The procedure: a laparoscopic cholecystectomy, meaning the removal of the gallbladder through small incisions. It is one of the most common procedures in the world, but also a gold standard for evaluating the surgical competence of a system. Each step was followed — trocar placement, dissection, clipping of the cystic duct, extraction.

The detail that matters: the surgeon was not at the patient's bedside. They piloted the two humanoids from a teleoperation console, much like flying a drone. The robots executed the movements on the anesthetized pig. The official EurekAlert press release specifies that this is the first time two teleoperated humanoids completed two surgeries during the same preclinical trial.

One point must be emphasized: this is not autonomous surgery. The robot makes no decisions. It is high-fidelity teleoperation. The distinction is crucial and too often blurred in mainstream reports.

Da Vinci vs humanoid: two radically different philosophies

The Da Vinci system, from Intuitive Surgical, has been the undisputed king of robotic surgery for over twenty years. But its approach is fundamentally different from what UCSD demonstrated.

Da Vinci is a dedicated system. Its arms are fixed to the patient, its instruments are specific to each type of procedure, and the surgeon is in the same room, back to the table, eyes glued to a 3D microscope. It is a tool for amplifying human capabilities, confined to an operating room.

The G1 humanoid used in San Diego is a generalist platform. Its arms are not surgical by design. The team adapted the interface, mounted compatible instruments, and demonstrated that a robot designed to walk and carry boxes could, with the right teleoperation, execute millimeter-scale movements.

This table does not say that the G1 is "better" than the Da Vinci. It says that the paradigm is different. The Boston Dynamics Atlas : le robot humanoïde qui fait tout seul shows the general trajectory of humanoids toward autonomy. In surgery, this autonomy is not the immediate goal — reliable teleoperation is.

The technique: teleoperation, latency, and dexterity

Surgical teleoperation is not new in itself. Pioneers like Jacques Marescaux had already performed transatlantic telesurgery (Lindbergh Operation) as early as 2001, using a dedicated robot and a dedicated very-low-latency line.

What changes with the humanoid is the very nature of the interface. The surgeon is not controlling articulated arms fixed to a robotic arm. They are controlling an entire body that stands up, orients itself, and positions its own arms relative to the patient. The kinematic complexity is of a completely different order.

According to the technical analysis by Ars Technica, the system relied on a force-feedback control loop (haptic feedback). The surgeon could feel the resistance of the tissues, which is essential to avoid unintentional injuries.

Latency remains the main bottleneck. In a local configuration (surgeon and robot in the same building), it is manageable. In an intercontinental configuration, each additional millisecond degrades precision and safety. The UCSD team did not publish exact latency figures for this trial, but the article in Nature notes that the performance was "consistent with the requirements of a standard laparoscopic procedure."

The dexterity of the G1's hands, compared to the articulated instruments of the Da Vinci, remains a point of friction. The fingers of a general-purpose humanoid are not designed to manipulate fragile hepatic tissue. Hardware adaptations — specialized forceps mounted on the end effectors — were necessary.

What this changes for global telesurgery

The central issue is not technical, it is geopolitical and health-related. A surgical system like the Da Vinci costs between $1.8 and $2.5 million, requires a dedicated room, trained staff, and a six-figure annual maintenance contract.

A $16,000 base humanoid, controllable remotely, changes the equation. Not tomorrow, not in 2027, but the trajectory is set. Michael Yip, a robotics expert at UCSD, publishes a perspective on July 9, 2026, in Science where he argues that surgical humanoids could solve the problem of access to care in underserved areas.

Imagine: a surgeon in Paris controls a humanoid in a clinic in the Sahel. The robot is delivered in a trunk, deployed by a local technician, and the specialist operates remotely. No need for a highly equipped operating room, just a reliable connection and a robot.

The obstacles remain colossal. Network latency in rural Africa, cross-border medical regulation, legal liability in the event of complications, the training of local teams for per-operative emergencies. But the physical prototype now exists. The rest is a matter of engineering and governance.

On the question of governance, in fact, the G7 Évian : Altman, Amodei et Hassabis réunis pour la première fois au sommet — et les États-Unis bloquent toute gouvernance contraignante illustrates the difficulty of obtaining an international framework, even for issues less critical than remote surgery.

Autonomy vs teleoperation: don't confuse everything

An essential point of clarification. Several media outlets, including TF1 Info and Fréquence Médicale, simultaneously reported advances in autonomous surgery — a robot trained on videos of procedures that performs gestures without real-time human control.

These two news stories are distinct. The UCSD trial published in Nature involves teleoperation: the surgeon controls every movement. The autonomy experiments reported by TF1 fall under a different paradigm, where the AI model learns gestures from data and reproduces them.

The two trajectories will likely converge. A humanoid might one day combine human teleoperation for critical phases and autonomous execution for routine gestures (opening, closing, irrigation). But in July 2026, the San Diego trial is strictly teleoperated. Any interpretation to the contrary is factually false.

Caution is all the more necessary as cybersecurity issues become central. A robot connected to a network for teleoperation is an attack surface. The news about IA auto-réplicante : pour la première fois, des modèles piratent des ordinateurs et se copient sur le réseau reminds us that networked autonomous systems present risk vectors that are still poorly understood.

The ethical issues no one is asking yet

The bioethics community is just beginning to process the arrival of LLMs in healthcare. The surgical humanoid adds a layer of complexity.

First question: liability. If a remote surgeon pilots a humanoid that causes an injury, who is responsible? The surgeon? The robot's manufacturer? The local team that prepared the patient? The hospital that authorized teleoperation? The current legal framework is not designed for this.

Second question: the dehumanization of care. The surgeon-patient relationship, even in conventional laparoscopic surgery, involves a physical presence. The patient knows that someone is there, in the room. With humanoid tele-surgery, this presence disappears. The psychological impact on the patient has not been studied.

Third question: equity of access. If humanoid tele-surgery develops, do we risk seeing surgeons from rich countries operating remotely in poor countries, without ever training local surgeons? This would be the extension of medical neocolonialism under the guise of technological innovation.

These questions have no answers today. They must be asked now, before the technology becomes irreversible.

Realistic timeline to human trials

Many headlines speak of a "world first" without specifying the context. Let's set the record straight with the necessary rigor.

This timeline is optimistic. It assumes that no major incidents occur during preclinical trials, that regulations adapt, and that funding follows. Historically, the Da Vinci took more than ten years between its first prototype and its widespread clinical adoption. The humanoid starts with an advantage (the platform already exists), but it also starts with a disadvantage (it is not a medical device by design).

Michael Yip's statements in Science are measured. He sees humanoids as a "potential solution," not an immediate one. Media outlets talking about "robot surgery in 2 years" are misleading their readers.

The role of AI in this revolution

The UCSD trial is primarily a mechatronics and teleoperation feat. But AI is underlying at several levels, and its role will grow.

First, movement stabilization. A generalist humanoid trembles more than a dedicated Da Vinci arm. Filtering and prediction algorithms, likely based on advanced control models, smooth the surgeon's movements to transmit them to the robot with superior precision.

Next, contextual assistance. Eventually, a system like GPT-5.5 or Claude Opus 4.7 could analyze the video feed in real time and alert the surgeon if they are approaching a critical structure (cystic artery, common bile duct). This is not surgical autonomy, it is decision support — comparable to a copilot in an airplane.

Even further, learning from demonstration. If thousands of teleoperated surgeries are recorded, a model could learn gesture patterns and propose partial automations. Gemini 3 Pro Deep Think or DeepSeek V4 Pro, with their reasoning capabilities, could be candidates for this type of multimodal analysis.

But here again, caution is required. No current AI model is certified for any real-time medical decision. And the distance between "analyzing a surgery video" and "assisting a surgical gesture in real time" is immense.

❌ Common mistakes

Mistake 1: Confusing teleoperation and autonomy

The UCSD trial is 100% teleoperated. The surgeon controls every movement. Writing that "the robot operated alone" is factually incorrect. The distinction is not academic — it changes the entire regulatory and ethical framework.

Mistake 2: Extrapolating directly to humans

A pig is not a human. The abdominal geometry differs, tissue reactions differ, potential complications differ. The trial is preclinical. Talking about "imminent clinical application" is irresponsible.

Mistake 3: Directly comparing G1 and Da Vinci on safety

Da Vinci has 20 years of clinical track record, thousands of documented procedures, and proven regulations. G1 has two cholecystectomies on pigs. Comparing their "performances" as if they were on the same level is a methodological error.

Mistake 4: Ignoring the network latency issue

Remote telesurgery requires a reliable, redundant network infrastructure with guaranteed latency. This infrastructure does not exist in the majority of places in the world where telesurgery would be the most useful.

❓ Frequently Asked Questions

Did a humanoid robot really operate on a human?

No. The UCSD trial, published in Nature in July 2026, was conducted on living pigs in a preclinical setting. No human has been operated on by a humanoid to date.

What is the difference with the Da Vinci robot?

Da Vinci is a dedicated, fixed, expensive surgical system ($1.8-2.5 M), where the surgeon is in the same room. The G1 humanoid is a mobile, general-purpose, low-cost platform (~$16,000) that can be controlled remotely.

Did the robot decide on the movements to make?

No. It was pure teleoperation: the surgeon controlled every movement in real time from a console. No AI made any surgical decisions.

When are human trials scheduled?

No date has been set. With the necessary optimizations, additional preclinical trials, and the regulatory process (FDA/EMA), the most realistic timeline is 2029-2030 at the earliest for a Phase I.

What is the cost of a Unitree G1?

The base version is announced at around $16,000 (July 2026, check on unitree.com). The surgical adaptations (end-effectors, interface, force feedback) add a cost that is not public to date.

✅ Conclusion

A general-purpose humanoid performed complete teleoperated surgery on a living subject, and it is published in Nature. The rest — intercontinental telesurgery, real-time AI assistance, human trials — is still ahead of us. But the paradigm has changed: robotic surgery is no longer the exclusive domain of dedicated machines. It could become just one application among others for versatile humanoid platforms. If you work in medtech or robotics, follow this topic very closely.