On 25 September 2026, FDA’s Center for Devices and Radiological Health published draft guidance on robotically assisted surgical devices, the first time the agency has pulled its premarket expectations for this class into a single document. It covers the whole submission: device description, predicate comparison, non-clinical testing, software, cybersecurity, human factors, in vivo evaluation, and labelling. It is a draft, open for comment until 24 November 2026, and any manufacturer with a US filing on the roadmap should read it as a preview of how reviewers will think, not as a rulebook in force.
Key insight
For anyone filing robotically assisted surgical devices, Appendix A is the part to act on first. Data from one complex, high-risk umbrella procedure can support authorisation for lower-complexity covered procedures in the same surgical specialty, without separate procedure-by-procedure data. A team that picks a defensible umbrella procedure before designing its clinical plan carries a smaller evidence burden for the same indications. A team that picks it late pays for those indications twice.
What the Document Is, and What It Is Not
The document is a Level 1 draft guidance, issued under FDA’s good guidance practices regulation at 21 CFR 10.115, with its notice of availability at 91 FR 60967, Federal Register document 2026-19704, under Docket No. FDA-2026-N-9505. The issuing office is CDRH’s Division of General Surgery Devices, and the guidance carries document number GUI01500081. Level 1, in FDA’s taxonomy, means an original guidance rather than a revision.
Two caveats govern everything else in it. First, it is not binding: the guidance states that it “does not establish any rights for any person and is not binding on FDA or the public,” and a manufacturer may use an alternative approach that satisfies the applicable statutes and regulations. Even the word “should” carries a defined meaning here: suggested or recommended, but not required. Second, it is not final. Comments close on 24 November 2026, and every “should” in the text can shift before a final version exists. That is why the rest of this article reads the draft as a map of reviewer thinking.
| Pathway | Citation | Applies to |
|---|---|---|
| 510(k) | 21 CFR 807 Subpart E | Premarket notifications (with predicate comparison) |
| De Novo | 21 CFR 860 Subpart D | New device types without a predicate |
| PMA | 21 CFR 814 Subparts A–E | Class III devices |
| IDE | 21 CFR Part 812 | Clinical studies supporting the above |
Scope: The Exclusions Are the News
FDA’s definition will surprise no one in this field. A teleoperated, software-controlled system, used under the direct control or supervision of a qualified practitioner, on a leader/follower architecture, with limited automated functions. One boundary is worth noting: endoluminal use means existing luminal cavities, expressly not an intravascular approach. If you build robotically assisted surgical devices, the definition is not where your attention belongs. The exclusions are.
Four categories sit expressly outside that scope, and if your product is one of them, this document is not your submission roadmap:
| Out of Scope | Examples / Details |
|---|---|
| Pre-operative planning and stereotaxic systems | Orthopedic (OLO), neurological (HAW), ENT (PGW) stereotaxic instruments — 21 CFR 882.4560 |
| Interventional catheter systems | Devices assisting placement or use of interventional catheters, or navigation through blood vessels |
| Remotely teleoperated RASDs | Operator console not co-located in the same operating room as the bedside subassembly |
| Autonomous robots | Systems performing significant operative steps independent of a qualified practitioner |
Some of the non-clinical recommendations may still be relevant to remote and autonomous systems, but the submission framework does not apply to them. And if there is genuine doubt about whether a device counts as a RASD, the guidance’s answer is to ask FDA through the Q-Submission Program.
Appendix A: The Clinical Evidence Shortcut
Appendix A is the most commercially significant part of the guidance because it lets one body of clinical evidence do the work of several. The mechanism is a pair of definitions.
An umbrella procedure is one of sufficient complexity and risk that its performance data can be leveraged to support authorisation for other procedures. A covered procedure is one in the same surgical specialty, of lesser complexity and risk, that leans on that data. In FDA’s own words, additional in vivo performance data for a covered procedure may not be needed at all.
In practice, in vivo data from a high-complexity procedure such as a radical prostatectomy can support authorisation for lower-complexity urology procedures without procedure-by-procedure studies. The condition FDA attaches is that the benefit-risk profile of the umbrella procedure should be similar to or greater than that of the covered procedures.
Two constraints bound the paradigm. It supports adding specified surgical procedures to an authorisation, but not other changes to the indications for use, such as treating a specific disease or condition. And there is no fixed list of umbrellas and covered procedures. FDA provides illustrative tables by specialty, expects manufacturers to propose their own structure, and offers the Q-Submission Program to confirm that structure before the clinical plan is locked.
| Specialty | Illustrative umbrella procedure | Illustrative covered procedures |
|---|---|---|
| Urology | Radical prostatectomy | Simple prostatectomy, lymphadenectomy |
| Colorectal | LAR/TME | Colectomy variants, small bowel resection, lysis of adhesions, rectopexy, APR, appendectomy |
| Gastric | Roux-en-Y bypass | Gastrectomy, cholecystectomy, splenectomy, gastric sleeve, Nissen fundoplication |
| Gynecology | Radical total laparoscopic hysterectomy | Salpingectomy, oophorectomy, ovarian cystectomy, endometriosis resection |
| Thoracic | Lobectomy, lymphadenectomy | Wedge resection, segmentectomy |
| Cardiac | Mitral valve repair | Mitral valve replacement, ASD repair, tricuspid valve repair |
Because the choice is device- and specialty-specific, the illustrative tables are a starting point, not a menu. Designing the clinical plan around a defensible umbrella procedure is the highest-leverage decision a robotics team makes under this guidance, and it is worth a Q-Sub of its own.
What the RASD Submission Must Contain
Section IV is the core of the guidance for robotically assisted surgical devices, eighteen lettered parts running from device description through labelling, each telling a manufacturer what a reviewer expects to find at that position in the file. It opens where every reviewer opens, with the device description.
Device description and predicate comparison
Provide a whole-system overview first, then each subassembly: images, principle of operation, a kinematic diagram, a connections diagram covering how the subassemblies talk to each other and to external servers, and a data transmission representation. For a 510(k), the description must also carry a statement of similarity to and differences from the predicate, under 21 CFR 807.87(f) and section 513(i) of the FD&C Act.
The wording of that comparison has its own rule. The terms “identical” and “same” are reserved for features that are demonstrably identical: same manufacturer, same design, same materials, same functions. Anything short of that gets described as a difference, with a rationale for why it does not affect substantial equivalence. The guidance supplies worked comparison tables, and a summary table of new or altered features is recommended.
Accessories and compatible devices
Accessories and third-party devices get the same treatment: described in detail, listed in the labelling, with compatibility shown by performance testing of the combined system, adapters and drapes included.
RASD-specific non-clinical performance testing
The non-clinical performance section names six areas of system-level testing, each with its own acceptance logic:
- System teleoperation — total system latency within pre-defined, clinically justified limits under worst-case load; minimum sample size of three RASDs, or a justification if fewer
- System controls — integrated control loops under signal noise and unexpected impulse forces; safe-state transition; emergency and safety stops
- Kinematics and dynamics — singularities identified on the kinematic diagram and mitigated in software; accuracy, repeatability and resolution across three RASDs; workspace reachability across patient extremes; collision detection
- Visualisation and display — resolution, distortion, field of view, colour rendition, thermal safety (IEC 60601-2-18)
- Additional RASD features — risk assessment for haptics, eye tracking, integrated imaging, and AI/ML algorithms; PCCP for post-authorisation software updates
- Reliability — 95% reliability with 95% confidence for actuating instruments under worst-case loading and reprocessing; cantilever bending; failure thresholds
Software
Software is where the file gets thick. FDA generally expects RASD software to warrant an Enhanced documentation level; proposing Basic requires a scientific justification based on software risk. Verification and validation should run at unit, integration, and full-system level, with traceability linking requirements to design, testing, and risk controls across the lifecycle, and testing on the final commercial software version.
Cybersecurity
Cybersecurity is where it becomes statutory. If the device meets the definition of a cyber device under section 524B(c) of the FD&C Act, the submission must meet the premarket cybersecurity requirements of section 524B(b): penetration testing across every network element that communicates with the system, plus a scenario in which the RASD operates on a compromised hospital network.
Wireless and human factors
Wireless coexistence risk management should follow AAMI TIR69, with test methods from ANSI USEMCSC C63.27. Human factors validation must use the final interface, in a realistic simulated operating room, with the team that actually uses the robot: surgeon, bedside assistant, and scrub nurse working simultaneously, emergency scenarios included.
Clinical Data and the Nine Surgical Endpoints
At some point the robot has to work inside a living body, and Section IV.Q is where FDA says what that looks like. The logic is stepwise, what FDA calls the least burdensome approach: bench data first, non-clinical in vivo studies where bench testing cannot sufficiently support the submission, and clinical data where new technologies, interfaces, or user populations raise questions non-clinical models cannot answer.
Chronic animal studies should assess morbidity and mortality, wound healing, instrument effects such as vessel sealing, and the durability of reconstructive elements like suture lines and anastomoses. Every animal should receive a systematic necropsy, gross and histologic, performed by a board-certified veterinary pathologist, with histology on all tissues the device could have affected. Validated non-animal alternatives can be proposed through a Q-Sub.
Where clinical data is needed, FDA recommends assessing nine primary surgical endpoints, and the list tells you what the agency is actually authorising:
- Length of hospital stay
- Intraoperative adverse events
- Transfusion rate (estimated blood loss)
- Conversion rate
- Post-operative adverse events (through 30 days)
- Readmission rates (through 30 days)
- Reoperation rates (through 30 days)
- Mortality
- Operative time
Clinical studies for investigational RASDs are generally classified as significant risk under 21 CFR 812.3(m), which brings full Part 812 oversight, IRB review under Part 56, and informed consent under Part 50. Data collected outside the United States must meet 21 CFR 812.28, and must reflect US patient demographics, comorbidities, and clinical practice.
Labelling: The Statement FDA Wrote for You
Labelling must satisfy the statutory requirements of the pathway being used: 21 CFR 807.87(e) for a 510(k), 814.20(b)(10) for a PMA, 860.220(a)(18) for a De Novo, and 801.109 for prescription devices. The instructions for use should carry consistent indications, warnings and contraindications, prominently placed emergency withdrawal procedures, patient selection criteria, learning-curve information, reprocessing and sterility instructions, and EMC and cybersecurity guidance.
One labelling statement does the most work. To clarify that authorisation covers the execution of specific surgical procedures rather than the treatment of disease, the guidance provides the exact wording to include: “Outcomes related to the specific treatment of underlying disease or patient condition were not evaluated as part of the marketing authorization for this robotically-assisted surgical device. Clinical data for the authorization were based on evaluation of the device as a surgical system that assists in the accurate control and performance of coordinated surgical tasks in the form of specific surgical procedures.” A second labelling fact is shorter: absent testing that demonstrates otherwise, the system should be labelled “MR Unsafe”.
What This Means for a Robotics Team
For a team building robotically assisted surgical devices, the practical effect is that the clinical plan is now the critical path, not the bench file. The umbrella and covered paradigm rewards teams that pick a defensible umbrella procedure early, because one data set can carry several indications at once. The bench testing is demanding but bounded. The clinical evidence is where schedules and budgets actually break.
The cybersecurity file stops being a formality too. Penetration testing across every connected subassembly, plus a compromised-network scenario, is genuine engineering work, and section 524B makes it statutory rather than discretionary for cyber devices. Teams using third-party imaging, cloud analytics, or telemetry will find the obligation travels up the supply chain, into vendor contracts.
The cheapest risk reduction in the entire document is a conversation. FDA names the Q-Submission Program for devices near the scope boundary, for umbrella and covered proposals, for non-animal alternatives, and even for MR Conditional labelling. A pre-submission on scope and clinical structure costs a fraction of one wrong-docket cycle.
For teams whose first filing is elsewhere, in the EU under the MDR or in India under the Medical Device Rules 2017, this document is still worth the read. It is the clearest public statement of the technical depth a major regulator now expects from this device class, and every category in it reads like a preview of the questions any regulator of surgical robots is learning to ask. The same traps catch teams in every market.
The Traps This Guidance Will Catch
Treating it as final. It is a draft open until 24 November 2026. Design to the underlying regulations, 21 CFR 807, 814, and 812, and use the guidance as a preview of reviewer expectations.
Assuming autonomous and remote operation are covered. Both are expressly out of scope. The technical recommendations may still speak to those systems, but the submission framework does not apply, and FDA’s stated route for them is a Q-Sub.
Using “identical” loosely. The guidance reserves the word for demonstrably identical features, same manufacturer, design, materials, and functions. Over-claiming sameness in a substantial equivalence comparison invites the exact deficiency letter the comparison was meant to prevent.
Running human factors validation with a single user. The guidance expects the surgeon, the bedside assistant, and the scrub nurse, working simultaneously in a simulated OR, because that is how the device is actually used.
Leaving latency and singularities as afterthoughts. Total system latency, from the surgeon’s hand input at the console to instrument motion to visual feedback, needs pre-defined clinically justified limits, and singularity points must be identified on the kinematic diagram and mitigated in software. Both are named testing expectations in the guidance.
What to Do Before 24 November 2026
- Classify your device: in scope, or one of the four out-of-scope categories.
- Model your umbrella and covered structure using the Appendix A tables, and identify where your clinical plan changes.
- Gap your file against the six non-clinical test areas and the software documentation level expectation.
- Scope your cybersecurity testing, including penetration testing and the compromised-network scenario, and check vendor support.
- Decide whether to comment. If you have real RASD submission experience, the comment window is your only chance to shape the final guidance. Submit comments to Docket No. FDA-2026-N-9505 — submissions are public.
Frequently Asked Questions
Does this guidance apply to a device with no predicate?
Yes. It covers De Novo classification requests (21 CFR 860 Subpart D) and PMAs (21 CFR 814 Subparts A–E) as well as 510(k)s. The predicate-comparison section applies only to 510(k) submissions.
Is this guidance binding right now?
No. It is draft, not for implementation, and states that it is not binding on FDA or the public. You may use an alternative approach that satisfies the applicable statutes and regulations. The recommendations nonetheless signal how reviewers will assess RASD submissions once the guidance is finalised.
What happens if my device is out of scope?
FDA recommends engaging through the Q-Submission Program for device-specific feedback. Parts of the non-clinical recommendations may still be relevant to remote and autonomous systems, but the submission framework does not apply.
Preparing a first US submission for a surgical robot? See how Regulatory Frontier guides medtech teams through CDSCO, EU MDR, and FDA pathways.
Sources
FDA draft guidance, Robotically-Assisted Surgical Devices—Premarket Submissions (September 2026), CDRH, document GUI01500081. Federal Register notice of availability, 91 FR 60967 (25 September 2026), Docket No. FDA-2026-N-9505. 21 CFR 10.115, 807.87, 807.92, 882.4560, Part 812, Parts 50 and 56, 814.20(b)(10), 860.220(a)(18), 801.109. FD&C Act sections 513(i), 520(o)(2), 524B.