Conventional robotic surgery was conceived with one core purpose: enhancing the surgeon’s own abilities. Platforms such as da Vinci filter out hand tremor, scale movement, improve visualization and increase precision when working in confined spaces. The robot, however, remains a tool purpose-built for specific procedures, operating at all times under the surgeon’s direct control.
Humanoid robots introduce a different premise altogether. Their principal advantage need not be operating better than we do, but rather offering a generalist physical platform able to move through an environment built for people, handle our own instruments and interact with other professionals and equipment. Two recent papers published in journals of the Nature group approach this possibility from complementary angles: one demonstrates, experimentally, how far this technology has already progressed, while the other proposes how it should be introduced, gradually and cautiously, into an environment as demanding as the operating room (OR).
This question is particularly relevant to cardiac surgery. Few specialties bring together, within a single space, such a density of technology, personnel, devices and critical decisions. Precisely for that reason, it may be worth separating two issues that are often conflated: whether a humanoid robot might one day operate, and whether it could prove useful long before it ever does.
Liang et al. developed a laparoscopic teleoperation system built on the commercial humanoid robot Unitree G1, adapted to handle conventional laparoscopic instruments through a console equipped with stereoscopic vision. The system was evaluated in successive stages, beginning with benchtop studies, followed by laparoscopic skills testing, and finally an in vivo porcine model.
The technical tests showed that the humanoid could work within a surgical workspace compatible with certain trocar placements, though with notable limitations. Latency between the operator’s movement and its robotic execution was approximately 156 ms, and precision was especially poor during curved motions. In the skills tests, the humanoid outperformed manual surgery on certain parameters but remained clearly behind the da Vinci Xi platform on the more complex tasks.
Finally, an experienced surgeon performed two porcine laparoscopic cholecystectomies using the teleoperated humanoid. Both procedures were completed without conversion, including dissection, identification of anatomical structures and separation of the gallbladder from the hepatic bed. Clipping of the cystic duct and ligation of the cystic artery, however, were carried out with conventional laparoscopic instruments, since no clip appliers compatible with the robot were available, after which robotic control resumed to complete gallbladder mobilization. Several interruptions were nonetheless required for recalibration and repositioning, and problems with reach, force, latency, overheating and maintenance of sterility were identified.
In a complementary contribution, Kernbach et al. propose a staged model for integrating humanoid robots into the operating room. Following a preclinical validation phase, they suggest starting with routine, low-risk activities, subsequently advancing toward support tasks close to the surgical field, and reserving higher-risk intraoperative functions for later stages, to be undertaken only once safety, reliability and clinical benefit have been demonstrated.
The authors of the experimental study conclude that present-day humanoid robots have reached a level of development sufficient to demonstrate the technical feasibility of laparoscopic surgical manipulation in vivo, but that they retain substantial limitations in precision, reach, latency, stability, sterility and integration with surgical workflow that rule out clinical use in the near term.
Kernbach et al. hold that their incorporation into the operating room should occur progressively, beginning with peripheral, low-risk tasks and expanding their responsibility only after safety has been demonstrated at each level. The authors understand that, at least initially, humanoids should be conceived as collaborators within the surgical team rather than as autonomous replacements for its professionals.
COMMENTARY:
There is something genuinely striking about the work of Liang et al.: a commercial humanoid robot never designed for surgery proved capable of picking up laparoscopic instruments built for human hands and taking part in two complete surgical procedures on living animals. A few years ago, this scene would have belonged more to science fiction than to the pages of Nature. The advance should not, therefore, be underestimated.
Nor should it be overstated. The humanoid used in the study is not an autonomous surgeon. The surgeon continues to see, interpret and decide from a console while the robot reproduces those movements. Even in a relatively standardized cholecystectomy, recalibration and repositioning were frequent, and human assistance remained necessary throughout much of the procedure. The gap relative to a platform purpose-built for surgery also remains considerable: the da Vinci Xi was faster, more precise and more ergonomic, whereas the humanoid showed greater latency, movement errors, reach limitations and a heavier cognitive load for the operator.
A direct comparison between the two systems, however, is probably not the most interesting question. da Vinci represents the evolution of a specialized tool designed to optimize one particular surgical act. The humanoid is pursuing something different: becoming a generalist platform capable of performing many different tasks within an environment and with instruments originally designed for people. Its potential advantage would therefore not lie in suturing a coronary vessel better than a robot purpose-built for that task, but in being able to do one thing now and something entirely different a few minutes later.
This distinction becomes especially relevant in cardiac surgery. Picturing a humanoid autonomously performing a coronary anastomosis, a mitral repair or controlling unexpected bleeding today is difficult to reconcile with the capabilities demonstrated in these studies. Millimetre-level precision, the need to interpret deformable tissue, the requirement for an immediate response to unforeseeable situations and, above all, the consequences of a single error make our specialty probably one of the most demanding possible settings for robotic autonomy.
Paradoxically, that same cardiac operating room offers a host of applications that are far closer to realistic implementation. Preparing or bringing over supplies, transporting devices, locating instruments, performing counts, handling non-sterile systems, holding cameras or retractors, assisting with instrument changes or carrying out certain repetitive tasks are technically less ambitious functions, but potentially far closer to real-world deployment.
Cardiopulmonary bypass offers an even more telling example. It seems unreasonable to suggest that a humanoid could replace the perfusionist, because this professional’s essential function does not consist of moving tubing or operating controls, but of continuously interpreting a changing physiological state and responding to unexpected events. What can be imagined, however, are systems capable of providing physical assistance with certain tasks, checking supplies, integrating information from different devices, detecting protocol deviations or carrying out predefined actions under human supervision. Something similar could apply to ventricular assist devices, ECMO or the many other devices that are now a routine part of a modern cardiovascular operating room.
This is probably where the most interesting contribution of the Kernbach et al. paper lies. The path toward what is termed embodied AI need not begin by replacing the professional who performs the most complex task. It can begin by relieving that professional of the repetitive, physical or logistical functions that consume time and attention. The value of the humanoid would initially lie in augmenting the team’s capabilities, not in replacing them.
This strategy also helps resolve an apparent paradox. A robot with legs, arms and a human-like appearance can look needlessly complex for transporting supplies or operating a device. For many tasks it probably is, and a simpler mobile platform would be cheaper, more stable and safer. The real case for the humanoid would only emerge if a single machine manages to perform many different functions without substantially altering an operating room built around human anatomy and movement. Its economic and clinical justification will therefore depend less on how spectacular any single demonstration looks and more on the versatility it can deliver across an entire surgical day.
The limitations of the current evidence, in any case, call for a measure of caution. The in vivo study included only two animals, a single lead surgeon, no randomization, numerous interventions by the human team and technology that remains experimental. Having completed two cholecystectomies demonstrates feasibility, not clinical safety. We also do not yet know how reliable these systems would be over prolonged procedures, how they would behave in the face of unexpected failures, what their real cost of implementation would be, or whether they can be integrated without disrupting the team dynamics that are essential to surgery. Even something as seemingly elementary as guaranteeing sterility remains unresolved.
There is, moreover, a fundamental difference between teleoperation and autonomy. The Liang study demonstrates, above all, mechanical capability: the humanoid can physically execute certain surgical actions. The genuinely difficult leap will be equipping it with the capacity to understand context, anticipate events, decide what to do and recognize when it should not act. That challenge belongs less to mechanical robotics and much more to artificial intelligence, safety, regulation and professional accountability.
For that reason, these studies probably do not herald the imminent arrival of the “robot surgeon.” What they point to may be more interesting still: the emergence of a new kind of technological member of the surgical team. In cardiac surgery, a field that has spent decades progressively incorporating perfusion, assist devices, navigation, advanced imaging, structural intervention and robotics, it would not be surprising if the next shift affected not only how we operate, but who, or what, works alongside us.
Perhaps, then, the right question is not when we will see a humanoid perform a cardiac operation. We should first ask which tasks in our operating room we still carry out today simply because, until now, only a person could do them. It is quite possible that this is where this revolution truly begins.
REFERENCES:
Liang Z, Thareja N, Zhang P, Joyce C, Atar S, Richter F, et al. In vivo feasibility study of humanoid robots in surgery. Nature. 2026. doi:10.1038/s41586-026-10796-x.
Kernbach JM, Brat G, Topol EJ, Rajpurkar P. Humanoid robots in the operating room: a framework for staged integration of embodied AI in surgery. NPJ Digit Med. 2026 Jul 13. doi: 10.1038/s41746-026-02853-6.
