Surgical Robotics: Systems, Safety and Autonomy

Surgical robotics is the one branch of robotics that cleared the hardest deployment bar decades ago. An operating room is an unstructured, deformable, sterile environment where a fault can kill a patient in seconds and a defect report goes to a federal regulator. Into that room, Intuitive has placed 11,106 da Vinci systems as of the end of 2025, and its fifth-generation platform shipped with a force-sensing capability the company describes as a first for any surgical technology in any modality Intuitive Surgical 2026 Intuitive Surgical 2024 . None of it is autonomous in the way this wiki's manipulation modules mean the word. Every da Vinci motion is a surgeon's command, scaled and tremor-filtered. That combination, mass deployment plus near-zero autonomy, is not a paradox. It is the honest answer to the precision and reliability bar this module describes, and it sets a calibration for every autonomy claim elsewhere in robotics.

The comparison that matters is with the learned-manipulation program this wiki spends most of its length on. Research policies report 80 to 90 percent task success on bench tasks and are called promising; a surgical device must not injure the patient, ever, in a domain where success rate percentages of that size would be a catastrophe. The field reached deployment anyway, and it did so by changing what the robot is asked to do: not decide, but execute the surgeon's decision more precisely than a hand can. Read that as a strategy, and the rest of surgical robotics, from challengers like CMR and Moon Surgical to the research results in supervised autonomous suturing, falls into place as a set of bets on how far up the autonomy ladder that strategy can be pushed before the bar blocks it.

The teleoperation baseline

The da Vinci architecture is the pattern the whole field inherited. A surgeon sits at a console away from the table and manipulates hand controllers; a patient-side cart holds three or four articulated arms carrying wristed instruments and an endoscope; a vision stack delivers a magnified stereoscopic view. The machine contributes what a hand holding a laparoscopic stick cannot: motion scaling that divides hand movement into finer instrument movement, tremor filtering, and wrists that articulate inside the body rather than at its wall. The surgeon remains the planner and the controller. In the framework Yang and colleagues proposed for the field, that is Level 0 autonomy, no autonomy, and the framework's authors are explicit that a surgical robot with motion scaling still belongs there because the output represents the surgeon's desired motion Yang 2017 .

Three decades of iteration on that baseline produced the installed base the challengers now compete against. Da Vinci 5, cleared in March 2024 as a fifth-generation multiport system, carries more than 150 enhancements over Xi, including improved controllers with vibration and tremor control, a next-generation 3D display, integrated insufflation and electrosurgery, and more than 10,000 times the computing power of its predecessor Intuitive Surgical 2024 . The most technically consequential addition is Force Feedback: optional instruments that measure and let the surgeon feel the subtle forces the tips exert on tissue. In Intuitive's preclinical trials across experience levels, the capability reduced force exerted on tissue by up to 43 percent Intuitive Surgical 2024 . The commercial gravity of the platform comes through in its deployment figures: the installed base grew 12 percent during 2025 to those 11,106 systems, da Vinci procedures grew roughly 18 percent worldwide in 2025, and 303 of the 532 systems placed in the fourth quarter were da Vinci 5s Intuitive Surgical 2026 .

Surgical robotic systems compared: Intuitive da Vinci, CMR Versius, and Moon Surgical Maestro, by design focus, differentiator, and regulatory path, with each system’s position on the six-level autonomy scale of Yang et al.
System	Design focus	Differentiator	Regulatory path
Intuitive da VinciL0	Multiport and single-port teleoperated soft-tissue surgery; the incumbent platform	da Vinci 5 ships 150+ enhancements over Xi, including Force Feedback instruments and 10,000x the computing power	Cleared since 2000; da Vinci 5 via 510(k) in March 2024, 11,106 systems installed as of end-2025Level 0: every motion is the surgeon’s command, scaled and tremor-filtered
CMR VersiusL0	Compact, modular multiport system that moves between operating rooms and care settings	Biomimetic arm design with small fully wristed instruments and an open console; Versius Plus adds a boom and integrated insufflation	US entry via FDA De Novo (October 2024) for adult cholecystectomy; Versius Plus 510(k) December 2025Level 0: teleoperated, with the surgeon at an open console rather than enclosed
Moon Surgical MaestroL1	Two-armed hold-and-position assistant for standard laparoscopic surgery	Works with standard laparoscopic cameras and instruments; ScoPilot adds AI camera-following of the instrument tip (March 2025)	First 510(k) December 2022; commercial system cleared June 2024Level 1: robot assistance; ScoPilot automates camera positioning on demand while the surgeon keeps continuous control

The table's last column is where the three systems genuinely differ, and the differences are narrower than their marketing suggests. Every one of them is a Level 0 or Level 1 machine on the Yang scale.

The challengers and what they compete on

CMR Surgical's Versius attacks the incumbent on physical and economic footprint rather than capability. It is a modular multiport system whose arms are built as separate bedside units that can be wheeled between operating rooms, with an open console the surgeon stands or sits at, and instruments designed to biomimic the human arm for flexible port placement CMR Surgical 2024 . Its US regulatory entry is itself a datum about the field's conservatism: the FDA granted marketing authorization through the De Novo pathway in October 2024 for a single indication, adult cholecystectomy, the routine gallbladder removal that is soft-tissue surgery's standard entry point, making Versius the first multiport soft-tissue general surgical system authorized through that route CMR Surgical 2024 . The company paired the announcement with a figure that explains the whole segment's economics: of nearly 10 million annual major operating-room procedures in the US, only about 2.5 percent were robot-assisted CMR Surgical 2024 . The market argument for every challenger is the 97.5 percent, not the installed base it must displace.

Moon Surgical's Maestro is a different bet about where value sits. It is not a console-teleoperated surgery platform at all: a two-armed assistant robot positioned opposite the surgeon, it holds and positions the laparoscopic camera and instruments during standard laparoscopic procedures, works with the ordinary cameras and graspers a hospital already owns, and received its first 510(k) in December 2022 on the basis of substantial equivalence to a 1990s endoscope-positioning device Parra 2023 . Its clinical case is staffing: the system supported a 30-patient series of cholecystectomies, bariatric procedures, hernia repairs, colorectal cases, and antireflux operations performed by a single surgeon without an assistant Parra 2023 . In 2025 it gained the feature that makes it interesting to an ML engineer: ScoPilot, an AI camera-control capability cleared in March 2025, runs on NVIDIA Holoscan compute locally on the robot and lets the laparoscope follow the instrument tip on demand while the surgeon's hands stay on the instruments Moon Surgical 2025 . That is Level 1 robot assistance on the Yang scale: the machine automates one continuously supervised task, the camera, and nothing else Yang 2017 .

The precision and reliability bar

Why does a field with three decades of shipped systems still ship Level 0? The question has a precise answer, and it is not that the research does not exist.

The reliability arithmetic is unforgiving in a way manipulation researchers will recognize from this wiki's module on the reliability gap. A learned policy that succeeds at 95 percent per step compounds to 21.5 percent over 30 steps, which is a discouraging bench result in a lab and an unthinkable one across 2.5 percent of ten million annual procedures CMR Surgical 2024 . The failure mode is also asymmetric: an error that damages tissue is not a failed trial but an adverse event, reported to a regulator, with the manufacturer's name on it. Intuitive's own launch discipline shows the required posture: da Vinci 5 reached only a small number of US customers at first, deliberately, to generate data on real use before broader introduction Intuitive Surgical 2024 . Phased rollout under regulatory surveillance is the surgical equivalent of the evaluation discipline this wiki argues manipulation lacks.

Precision, by contrast, is the part of the bar the field cleared first. Motion scaling, tremor filtering, and wristed instruments give sub-millimeter effective precision inside a body, and the STAR result demonstrates that machine precision can beat expert hands outright. In the study most often cited for autonomous surgery, the Smart Tissue Autonomous Robot (STAR) performed supervised autonomous soft-tissue anastomosis in vivo, with the robot planning and executing suture placement while a surgeon supervised, and its outcomes on suture spacing and leak pressure matched or exceeded both expert surgeons and conventional robot-assisted technique in porcine studies Shademan 2016 . That is a genuine Level 2 task-autonomy result: the robot performed the suturing task itself, in living tissue, not in simulation. It is also a decade old, and no commercial system ships it. The gap between that demonstration and what is on the market is the reliability bar working as designed, not a technology gap.

The regulatory mechanism behind that conservatism is worth naming because it is the part robotics engineers least often model. A device is authorized for an indication, a named procedure in a named population, and expansion means a new clearance. The Yang editorial frames the deeper problem: at high autonomy levels the system stops being only a medical device and starts practicing medicine, which in the US falls to medical societies rather than the FDA, and no framework exists for licensing a robotic surgeon the way human surgeons are licensed Yang 2017 . The same editorial observes that the cost and time gap between the 510(k) and premarket-approval pathways is on the order of tens of millions of dollars and years of review, and that risk tolerance to autonomous robots is expected to change as autonomous machines such as self-driving cars become commonplace Yang 2017 . None of the three companies in this module is fighting that fight; they are competing on footprint, workforce, and workflow, which is where the money and the clearances are.

What robot learning takes from it

Two transfers matter for an ML engineer reading the rest of this wiki.

First, surgical robotics is the existence proof that verified reliability, not capability, gates deployment in the highest-stakes domains. The entire field is organized around a human-in-the-loop architecture whose every increment of automation is individually cleared: force sensing first as a display, camera control as an on-demand assistant, one procedure at a time. Compare that with the AV module's account of how driving autonomy is earned through statistical crash-rate evidence over hundreds of millions of miles, and with the drones module's hardware-identical race wins: same problem, three different evidentiary bars. The lesson generalizes to manipulation: the bar your policy must clear is set by the worst outcome of the task, not by the median outcome of the benchmark.

Second, the field is building exactly the data substrate learned policies need, whether or not that is anyone's stated goal. Da Vinci 5's force-sensing instruments add a measured interaction-force stream to surgical data science Intuitive Surgical 2024 , and Maestro's ambient sensing plus onboard compute generates multimodal kinematics and video that Moon Surgical explicitly frames as the substrate for next-generation physical AI capabilities Moon Surgical 2025 . Instrumented teleoperation recording what expert surgeons actually do is the same strategic asset this wiki's teleoperation module describes for manipulation data collection: the demonstrations are the training set. The loop is closing in the direction the manipulation literature predicts, from human demonstration toward assistance toward supervised autonomy, but at the pace the reliability bar allows, one cleared indication at a time. Where the frontier manipulation programs attempt to jump the bar with generalist policies, surgical robotics walks through it, and the field that walks is the one already inside. The one domain that raises the bar even higher, because the machine cannot be retrieved when it fails and the operator is minutes away at best, is space robotics.

Surgical Robotics

The teleoperation baseline

The challengers and what they compete on

The precision and reliability bar

What robot learning takes from it

See also

Linked from

References