Imitation Learning and Hierarchical Control for Surgeon-Independent Laparoscopic Procedures

Autonomous Soft Tissue Surgical Systems with Learned Dexterity

1. Problem Statement

Approximately 50 million surgical procedures are performed in the United States each year. The Agency for Healthcare Research and Quality estimates the annual cost of preventable adverse events in hospitalized patients at $17.1 billion, with post-surgical complications constituting the largest component (Van Den Bos et al., "The $17.1 Billion Problem: The Annual Cost of Measurable Medical Errors," Health Affairs, 2011). A study published in Annals of Surgery found that 1 in 10 patients who died within 90 days of surgery did so because of a preventable medical error (Zegers et al., 2009).

Surgeon-dependent variability is a root cause. In laparoscopic cholecystectomy — the most common intra-abdominal operation in the US, with over 1.2 million cases performed annually (StatPearls, NCBI Bookshelf) — bile duct injury rates have not declined significantly over the past 30 years despite advances in instrumentation. Conversion from laparoscopic to open procedure occurs in 1-10% of cases, increasing hospital charges from an average of $23,946 to $32,446 per case (Giger et al., "Laparoscopic cholecystectomy: what is the price of conversion?," Surgical Endoscopy, 2012). Suture consistency — spacing, depth, and tension — directly affects anastomotic leak rates and healing outcomes, yet varies substantially between surgeons and across procedures performed by the same surgeon under fatigue.

The fundamental limitation of current teleoperated surgical robots (da Vinci, Versius, Hugo RAS) is that they amplify the capabilities of a present, attentive surgeon but cannot compensate for human variability, fatigue, or the global shortage of experienced surgeons. The World Health Organization estimates that 5 billion people lack access to safe, affordable surgical care, with the disparity most acute in sub-Saharan Africa and South Asia where surgeon-to-population ratios are 100-fold lower than in high-income countries.

The unmet market need is a surgical system capable of executing defined procedural phases — tissue manipulation, suturing, clipping, cutting — autonomously under human supervision, with mechanical consistency that equals or exceeds expert surgeon performance regardless of operator fatigue, geographic location, or institutional surgical volume.

2. State of the Art

Three parallel research programs have converged to establish the technical feasibility of Level 4 autonomous surgery — systems that execute surgical tasks independently while a supervising surgeon monitors and can intervene.

Autonomous suturing with near-infrared tracking. The Smart Tissue Autonomous Robot (STAR) program at Johns Hopkins University, led by Axel Krieger, has progressed through three generations of autonomous soft tissue surgery systems. The foundational work (Shademan et al., Science Translational Medicine, 2016) demonstrated supervised autonomous intestinal anastomosis in live porcine models using plenoptic 3D imaging and near-infrared fluorescent markers for real-time tissue tracking. The system's suture consistency — spacing and bite depth — exceeded expert surgeon performance on multiple metrics. The work was supported by NIH NIBIB awards R01EB020610 and R21EB024707.

Autonomous laparoscopic surgery. Saeidi et al. (Science Robotics, 2022) advanced STAR to laparoscopic operation — the first autonomous robotic system to perform minimally invasive surgery on soft tissue. The system completed intestinal anastomosis in live porcine models with 83% of sutures placed autonomously, demonstrating lower coefficient of variance in suture spacing and bite depth compared to both manual laparoscopic and robot-assisted techniques performed by experienced surgeons. Anastomoses survived one-week in vivo without complication.

Hierarchical imitation learning for multi-step procedures. The SRT-H framework (Krieger et al., Science Robotics, 2025) demonstrated autonomous execution of the clipping-and-cutting phase of laparoscopic cholecystectomy — a multi-step sequence of 17 distinct surgical tasks including duct identification, clip placement, and vessel transection — in ex vivo porcine gallbladders. The system achieved 100% task completion across 8 unseen specimens without human intervention. Trained on approximately 18,000 demonstrations collected from over 30 porcine procedures, SRT-H uses a hierarchical architecture combining a language-conditioned high-level planner with a low-level motion policy, enabling real-time adaptation to anatomical variation. The system responds to spoken corrections during execution, enabling a supervising surgeon to guide behavior without physical takeover.

No commercial surgical robot operates at Level 4 autonomy. A systematic review published in npj Digital Medicine (Attanasio et al., 2024) found that all FDA-cleared surgical robots function at Level 1 (robot assistance, 86%) or Level 2-3 (task or conditional autonomy, 14%), with zero examples of Level 4 or Level 5 systems. The da Vinci 5 system, FDA-cleared in 2024, introduces force feedback and enhanced imaging but remains a teleoperated platform requiring continuous surgeon control.

3. Foundational Research

Shademan A, Decker RS, Opfermann JD, Leonard S, Krieger A, Kim PCW (2016). "Supervised autonomous robotic soft tissue surgery." Science Translational Medicine, 8(337), 337ra64. DOI: 10.1126/scitranslmed.aad9398. PubMed: 27147588. Demonstrated the first in vivo supervised autonomous soft tissue surgery using the STAR system. The system used plenoptic 3D and near-infrared fluorescent (NIRF) imaging to track tissue deformation in real time during intestinal anastomosis in porcine models. Results: autonomous suturing produced more consistent spacing (coefficient of variance 0.07 vs. 0.15 for expert manual technique) and higher leak pressure tolerance than both manual laparoscopic and robot-assisted approaches. This established that autonomous soft tissue surgery could achieve superior mechanical consistency compared to expert human performance in a clinically relevant procedure.

Saeidi H, Opfermann JD, Kam M, Wei S, Leonard S, Hsieh MH, Kang JU, Krieger A (2022). "Autonomous robotic laparoscopic surgery for intestinal anastomosis." Science Robotics, 7(62), eabj2908. DOI: 10.1126/scirobotics.abj2908. PubMed: 35080901. Advanced STAR to laparoscopic operation — the clinically dominant approach for abdominal surgery. The system performed end-to-end intestinal anastomosis in live porcine models, placing 83% of sutures autonomously with the remaining 17% requiring minor human adjustment for needle re-grasping. Suture spacing coefficient of variance was 0.08 (autonomous) vs. 0.14 (expert manual laparoscopic), with comparable leak pressures. Animals survived one week post-procedure without anastomotic complications. This is the first demonstration of autonomous minimally invasive soft tissue surgery in a living animal model.

Krieger A et al. (2025). "SRT-H: A hierarchical framework for autonomous surgery via language-conditioned imitation learning." Science Robotics, 10(104), eadt5254. DOI: 10.1126/scirobotics.adt5254. PubMed: 40632876. Demonstrated autonomous execution of a multi-step surgical phase — clipping and cutting of the cystic duct and artery during laparoscopic cholecystectomy — across 17 distinct tasks on 8 unseen ex vivo porcine gallbladders with 100% task completion. The system was trained on 18,000 demonstrations from 34 porcine procedures using a hierarchical architecture: a language-conditioned high-level planner selects surgical actions, and a low-level imitation learning policy generates instrument trajectories. The framework accepts spoken corrections during execution, enabling real-time human guidance without physical takeover. This represents the highest level of demonstrated surgical autonomy in a clinically realistic multi-step procedure.

Attanasio A, Scaglioni B, De Momi E, Fiorini P, Valdastri P (2024). "Levels of autonomy in FDA-cleared surgical robots: a systematic review." npj Digital Medicine, 7, 104. DOI: 10.1038/s41746-024-01102-y. PubMed: 38671232. Analyzed 53 FDA-cleared surgical robotic systems using the six-level autonomy framework (Level 0-5). Findings: 86% operate at Level 1 (robot assistance under continuous surgeon control), 14% at Level 2-3 (task or conditional autonomy), and zero at Level 4 or Level 5. Classification was predominantly via 510(k) (83%), with a growing number via De Novo. The review establishes that the regulatory pathway for higher autonomy levels remains undefined and will likely require either De Novo classification or PMA, creating both a barrier and a competitive moat for the first entrant.

Webster RJ et al. (2024). ARPA-H ALISS Program Award, Vanderbilt University. Award amount: up to $12 million. Project period: 2024-2027. ARPA-H's Autonomy at a Less Invasive Scale in Surgery (ALISS) program, awarded to a consortium led by Robert J. Webster III at Vanderbilt University with collaborators from Johns Hopkins University, University of Utah, and University of Tennessee. The program targets fully autonomous tumor resection from the trachea and prostate within three years, initially demonstrated in simulated conditions. The award includes placement of Virtuoso Surgical Systems at three participating sites for AI/ML development. This $12M federal investment validates that autonomous surgery has transitioned from academic curiosity to a funded national research priority.

4. Competitive Landscape

Intuitive Surgical (Sunnyvale, CA). Market capitalization exceeding $180 billion. The da Vinci platform holds approximately 80% market share in robotic surgery, with over 9,000 systems installed globally. The da Vinci 5, FDA-cleared in 2024, adds force feedback and 10,000x computing power over the Xi generation but remains a Level 1 teleoperated system — the surgeon controls every instrument movement in real time. Intuitive has not disclosed work on Level 4+ autonomous capabilities. The company's business model depends on surgeon operators and per-procedure instrument sales, creating organizational inertia against autonomy that would reduce both surgeon dependence and instrument-per-case consumption.

Medtronic Hugo RAS (Dublin, Ireland). The Hugo Robotic-Assisted Surgery platform received CE marking in 2021 and is pursuing FDA clearance. Level 1 teleoperation. Medtronic's primary competitive strategy targets price parity with da Vinci through modular arm design and instrument reprocessing, not autonomy.

CMR Surgical Versius (Cambridge, UK). Modular surgical robot system with CE marking and market presence across Europe and Asia-Pacific. Level 1 teleoperation with port-placement guidance features (Level 2).

No commercial entity offers a Level 4 autonomous surgical system. The competitive landscape consists entirely of teleoperated platforms. The academic programs demonstrating Level 4 capabilities (Johns Hopkins STAR/SRT-H, Vanderbilt ALISS consortium) are research-stage without commercial entities. The gap between demonstrated research capability and commercial availability is wide and represents a first-mover opportunity for the entity that navigates regulatory classification and manufacturing scale-up.

5. Total Addressable Market

Bottom-up calculation (US laparoscopic surgery — initial indication: cholecystectomy):

Annual laparoscopic cholecystectomy procedures (US): 1,200,000 (StatPearls, NCBI Bookshelf)
Procedures at hospitals with robotic surgery infrastructure: approximately 40% = 480,000 (based on da Vinci install base)
Autonomous system per-procedure fee premium over manual laparoscopic: $3,500 (comprising $1,500 autonomous instrument kit + $2,000 platform usage fee, positioned below the $3,500-$4,500 per-case robotic surgery cost premium currently charged by Intuitive)
Initial SAM (cholecystectomy at robotics-equipped hospitals): 480,000 x $3,500 = $1.68 billion annually
Expansion to additional laparoscopic procedures (appendectomy: 300,000/year; hernia repair: 800,000/year; colorectal: 170,000/year): additional addressable volume of approximately 700,000 procedures at robotics-equipped hospitals
Expanded US TAM (laparoscopic procedures): (480,000 + 700,000) x $3,500 = $4.13 billion annually
Value-based savings: autonomous consistency reducing preventable adverse events by 20% across addressable procedures would eliminate approximately $2.4 billion in annual complications costs, supporting payer adoption and reimbursement negotiation

Top-down cross-check:

The global surgical robotics market was valued at $8.31 billion in 2025 and is projected to reach $12.83 billion by 2030, growing at 9.07% CAGR (Mordor Intelligence, "Surgical Robots Market," 2025). A parallel estimate from SkyQuest projects $27.14 billion by 2030 at 14.7% CAGR. Autonomous surgical systems as a premium subsegment capturing 15-20% of the projected 2030 market yields $1.9-$5.4 billion — consistent with the bottom-up estimate given the US-only scope.

Reimbursement pathway: Robotic-assisted laparoscopic cholecystectomy is reimbursed under existing CPT codes 47562 (laparoscopic cholecystectomy) and 47563 (with cholangiography), with no separate CPT code for robotic assistance. Medicare national average facility payment: $652-$709 per procedure. Autonomous systems would initially bill under the same codes with potential for a new technology add-on payment (NTAP) to capture the autonomous premium during the initial adoption period, transitioning to procedure-specific reimbursement as utilization data supports dedicated code assignment by the AMA CPT Editorial Panel.

SAM refinement: Initial deployment constrained to academic medical centers and high-volume surgical centers with existing robotic infrastructure. Estimated 300 qualifying sites performing an average of 500 qualifying laparoscopic procedures per year: 150,000 procedures x $3,500 = $525 million initial annual SAM, scaling as the autonomous system demonstrates safety and consistency across broader surgical populations.

6. Research Gap and Commercial Opportunity

Three specific gaps separate published research results from a deployable autonomous surgical product:

Gap 1: Regulatory classification for Level 4 autonomous surgical systems. All 53 FDA-cleared surgical robots identified by Attanasio et al. (2024) operate at Level 1-3. No regulatory precedent exists for a system that independently executes surgical tasks while a human surgeon supervises. The FDA's 2023 guidance on AI/ML-enabled medical devices addresses algorithm updates but does not specifically address surgical autonomy levels. The entity that establishes the regulatory template — whether via De Novo classification, PMA, or a novel framework developed in collaboration with CDRH — sets the regulatory standard that all subsequent entrants must follow. This is a 2-4 year process that creates a durable competitive moat.

Existing teleoperation companies have organizational reasons not to pursue Level 4 autonomy. Intuitive Surgical's business model generates recurring revenue from surgeon-dependent per-case instrument sales; autonomy that reduces surgeon dependence threatens this revenue model. Medtronic and CMR Surgical are still pursuing regulatory clearance for their Level 1 teleoperated systems and lack the research infrastructure for autonomy development. ARPA-H's ALISS program targets research demonstration, not commercialization.

Gap 2: Sensor-rich end-effectors manufactured at clinical volumes. SRT-H uses custom instrumentation with force sensing, multi-view cameras, and articulated graspers designed for research environments. Clinical deployment of 150,000-480,000 procedures per year requires single-use or limited-reuse sterile instruments manufactured with consistent quality under ISO 13485. Designing for manufacturability — material selection for sterilization compatibility, sensor integration in a cost-constrained form factor, automated assembly and packaging — is a manufacturing engineering problem that academic robotics labs neither prioritize nor possess the expertise to solve.

Gap 3: Safety architectures for human-supervised autonomous operation. Current demonstrations use research-grade safety monitoring. Clinical deployment requires: validated failure detection (instrument malfunction, tissue anomaly, anatomical variation beyond training distribution), graceful degradation to human control, real-time surgical scene understanding for situational awareness, and audit trails satisfying FDA quality system requirements. The safety architecture must be demonstrated across statistically significant patient populations and validated against defined failure modes — a systems engineering challenge distinct from the algorithm development demonstrated in published research.

Commercial thesis: Johns Hopkins has demonstrated that autonomous surgery works — better than expert surgeons by quantitative metrics. The academic program has neither the manufacturing capability, regulatory strategy expertise, nor incentive to commercialize. Intuitive has the manufacturing and regulatory capability but organizational disincentives to pursue autonomy. The company that integrates demonstrated autonomous surgical algorithms with manufacturable instrumentation and navigates Level 4 regulatory classification captures the autonomous surgery market before either academic programs or incumbent device companies can respond.

7. Comparable Funded Projects

ARPA-H ALISS Program (2024-2027). PI: Robert J. Webster III, Vanderbilt University. Award: up to $12 million. Multi-institution consortium (Vanderbilt, Johns Hopkins, University of Utah, University of Tennessee) targeting fully autonomous tumor resection from trachea and prostate using Virtuoso Surgical System platform. The ALISS acronym — Autonomy at a Less Invasive Scale in Surgery — establishes autonomous surgery as a named ARPA-H research priority, signaling sustained federal investment in this domain.

NIH NIBIB R01EB020610 and R21EB024707. PI: Axel Krieger, Johns Hopkins University. These awards funded the STAR and SRT-H programs that produced the three landmark publications (2016, 2022, 2025) in Science Translational Medicine and Science Robotics. The progression from R21 (exploratory) to R01 (significant) funding reflects NIH's assessment that autonomous surgery has demonstrated sufficient preliminary data to justify sustained multi-year investment.

NSF National Robotics Initiative (NRI-3.0). Program-wide funding across multiple surgical robotics projects. NRI-3.0 specifically includes "medical and healthcare robotics" as a priority area, with multiple active awards supporting surgical automation research including tissue manipulation, computer vision for surgical scene understanding, and sim-to-real transfer for surgical tasks.

DARPA ARM-S Program (Autonomous Robotic Manipulation - Software). While focused on general manipulation rather than surgery specifically, the ARM-S program funded foundational work on autonomous dexterous manipulation under uncertainty that directly informs surgical autonomy — particularly grasp planning, force-controlled interaction with deformable objects, and learning from demonstration.

8. Opportunity Assessment

TRL evidence chain: TRL 4 for the integrated autonomous surgical system. Component TRLs: autonomous suturing algorithm TRL 5 (demonstrated in vivo in live porcine models, Saeidi et al., 2022). Autonomous multi-step procedural execution TRL 4 (demonstrated ex vivo with 100% completion, SRT-H, 2025). Manufacturing-ready instrumentation TRL 2-3 (research-grade instruments only). Safety architecture TRL 2 (ad hoc monitoring in research settings). The rate-limiting component is the manufacturing and safety engineering, not the algorithmic capability.

Top 3 technical risks and mitigation:

Anatomical variation beyond training distribution. The SRT-H training set of 34 porcine specimens does not capture the full range of human anatomical variation — adhesions from prior surgery, variant cystic duct anatomy (present in 12-18% of patients), acute inflammation changing tissue planes. Mitigation: initial clinical deployment restricted to elective cholecystectomy in patients screened by preoperative imaging for standard anatomy. Anomaly detection trained on the same vision architecture flags out-of-distribution anatomy and transfers to human control. Expansion to variant anatomy as the training dataset grows through supervised clinical use. Risk level: moderate.
Regulatory uncertainty for Level 4 autonomy classification. No precedent exists for FDA classification of a system that autonomously executes surgical steps. CDRH may classify it as Class III (PMA required) rather than Class II (De Novo), substantially increasing regulatory timeline and cost. Mitigation: structured pre-submission engagement with CDRH's Division of Surgical Devices, framing the system as "surgeon-supervised autonomous assistance" — analogous to adaptive cruise control in automotive (Level 2-3 ADAS) — rather than "autonomous surgery." The ARPA-H ALISS program's explicit government endorsement of autonomous surgery research strengthens the regulatory narrative. Risk level: moderate-high.
Manufacturing cost for sensor-rich single-use instruments. Current research instruments are custom-fabricated and reused across procedures. Clinical deployment at scale requires either single-use sterile instruments (high per-unit cost, simpler regulatory path) or validated reprocessing protocols (lower per-unit cost, reprocessing validation adds regulatory complexity). Mitigation: initial launch as limited-reuse instruments (validated for 10 procedures) to amortize sensor costs while establishing reprocessing protocols. Component cost reduction through ASIC integration, roll-to-roll sensor fabrication, and automated assembly. Risk level: moderate.

Regulatory pathway: De Novo classification is the most probable pathway, as no predicate device exists for Level 4 autonomous surgery. The system would be classified as a Class II medical device with special controls. The AI/ML algorithm component falls under FDA's 2023 guidance on AI/ML-enabled Device Software Functions. If the algorithm adapts based on surgical experience (continuously learning), a Predetermined Change Control Plan (PCCP) would be required, specifying the types of modifications the algorithm can make post-market without requiring new clearance. An alternative strategy is to submit a locked algorithm (trained, then frozen for deployment), which simplifies the regulatory pathway but forecloses on-device learning. The locked approach is recommended for initial clearance, with PCCP-enabled adaptive capability added via supplemental submission after establishing a safety track record.

Estimated timeline: 12 months for pre-submission meetings and classification determination, 18-24 months for preclinical testing and pivotal study design, 12-18 months for pivotal study execution, 6-12 months for FDA review. Total: 4-5.5 years to De Novo authorization.

The De Novo classification, once granted, creates a regulatory moat: subsequent entrants can file 510(k) referencing the De Novo as predicate, but only after the first entrant has established the classification — a 4-5 year head start.

9. Team Requirements

Successful development and commercialization of a Level 4 autonomous surgical system requires three intersecting capabilities:

Surgical anatomy and experimental design expertise. Deep understanding of laparoscopic surgical anatomy (abdominal wall layers, Calot's triangle, cystic duct and artery variants, hepatic hilum anatomy) and surgical physiology (tissue response to manipulation, electrocautery effects, hemostasis mechanisms). Required for: defining clinically meaningful endpoints for autonomous system evaluation, designing preclinical validation protocols that satisfy FDA expectations, identifying anatomical variants that require out-of-distribution detection, and translating between engineering metrics (suture spacing coefficient of variance) and clinical outcomes (anastomotic leak rate, complication-free survival).

Machine learning for embodied manipulation. Imitation learning architectures (behavioral cloning, DAgger), hierarchical policy design, vision-language models for surgical scene understanding, sim-to-real transfer for tissue manipulation. Evaluation methodology for comparing autonomous versus surgeon-performed procedures across patient-specific anatomical variation. Scalable compute infrastructure for training policies on large-scale surgical video datasets. Required for: building and validating the autonomous control system that enables surgeon-independent procedural execution.

Manufacturing engineering for medical devices. Design for manufacturability (DFM) of sensor-integrated surgical instruments — force sensor integration, camera miniaturization, articulated joint mechanisms in sterilization-compatible materials (medical-grade stainless steel, PEEK, silicone). Quality systems (ISO 13485) including design controls, process validation, sterilization validation (EtO, gamma, or vaporized hydrogen peroxide), and biocompatibility testing (ISO 10993). Required for: bridging the gap between custom research instruments and production-volume medical devices — the specific failure point where most surgical robotics research stalls. Without manufacturing engineering from project inception, research instrument designs propagate into production-incompatible architectures that require costly ground-up redesign.