In vivo feasibility study of humanoid robots in surgery

Technical Summary: In Vivo Feasibility Study of Humanoid Robots in Surgery

Problem Statement
While recent advances in actuation, control, and learning have propelled humanoid robots toward real-world deployment, their application in high-stakes clinical environments remains unproven. Healthcare faces a widening gap between clinical workload and skilled labor availability. Although current hospital robotics focus on logistics, there is a critical need to determine if general-purpose humanoids can safely perform embodied, safety-critical tasks like surgery. Unlike purpose-built surgical platforms (e.g., the da Vinci Surgical System) which utilize mechanically constrained Remote Centers of Motion (RCM) and proprietary instruments, general-purpose humanoids must navigate human-designed operating rooms using standard manual instruments. It remains unclear whether contemporary humanoids possess the necessary precision, stability, and safety assurance to meet the rigorous demands of minimally invasive surgery (MIS).

Methodology
The authors developed a systematic evaluation framework to assess contemporary humanoid technology for laparoscopic tasks. The study utilized a Unitree G1 humanoid robot, representing a commercially available, general-purpose platform.

• System Architecture: A teleoperation framework was constructed using a mobile operating console equipped with a stereo headset, a Master Tool Manipulator (MTM), and a foot pedal. The system maps intuitive human hand motions to scaled instrument control via an inverse-mapping algorithm.
• Kinematic Control: To address the lack of mechanical RCM constraints in humanoids, the system employs a virtual RCM. The control algorithm treats the manual laparoscopic instrument as an extended kinematic chain. Using stereo vision and ArUco markers detected by the robot's head camera, the system localizes the trocar port and enforces the RCM constraint through software, allowing the robot wrist to pivot the instrument shaft while controlling the tip.
• Instrumentation: The study utilized commercially available, non-actuated wristed instruments (LivsMed) designed for human use. A custom mount allowed the robot to actively control the instrument's wrist degrees of freedom, while a servo-driven interface managed the gripper's closing action.
• Evaluation Phases:
  1. Benchtop Characterization: Assessed workspace accessibility, maneuverability, and motion accuracy (latency and tracking error) using motion capture systems.
  2. Dry-Lab User Studies: Conducted with 18 participants (6 surgeons, 12 novices) and 13 medical professionals (8 junior, 5 senior surgeons). Tasks included rubber ring transfer and Fundamentals of Laparoscopic Surgery (FLS) peg transfer. Performance was benchmarked against the da Vinci Research Kit (dVRK), da Vinci Xi, and manual operation.
  3. In Vivo Porcine Study: Two live laparoscopic cholecystectomies were performed in a surgical simulation center. One humanoid acted as the surgeon, while a second humanoid (in one case) or a human assistant managed camera control and retraction.

Key Results

• Benchtop Performance: The humanoid achieved a system latency of approximately 156 ms (follower-leader), which is within the range of active development but higher than the <150 ms often desired for clinical systems. Motion tracking showed millimeter-scale accuracy for straight lines (1.30 mm RMS deviation) but higher errors for curved trajectories (10.40 mm radial deviation), primarily attributed to calibration inaccuracies in the extended kinematic chain of manual tools.
• Dry-Lab User Studies:
  • Ring Transfer: The humanoid platform significantly reduced weighted error scores compared to manual operation for both novices and surgeons, performing comparably to the dVRK in error metrics but with longer completion times.
  • FLS Peg Transfer: The humanoid outperformed manual operation in both speed and error reduction, achieving a higher balanced FLS score. However, it remained significantly slower and less accurate than the da Vinci Xi system. Senior surgeons demonstrated the best performance across all platforms, achieving their lowest error rates on the humanoid.
  • Workload: Subjective feedback indicated that while the humanoid improved performance over manual operation, it required higher mental and physical demand compared to the da Vinci Xi, largely due to latency and reachability constraints.
• In Vivo Porcine Study: Both cholecystectomies were completed robotically without conversion to open surgery or conventional laparoscopy.
  • Efficiency: Active console time was efficient, with rapid execution of tissue dissection and critical view of safety.
  • Interruptions: The workflow included "major pauses" (>3 minutes) for robot redeployment and recalibration due to dynamic RCM shifts caused by animal respiration and base drift. "Micro-pauses" (<1 minute) for camera cleaning and instrument adjustment were comparable to human assistant tasks.
  • Outcomes: No major intraoperative complications occurred. Minor issues (biliary spillage, bleeding) in the second case were managed successfully.

Significance and Claims
The paper claims to provide the first evidence-based assessment of contemporary general-purpose humanoids performing in vivo laparoscopic surgery. The study demonstrates that:

1. Technical Feasibility: Humanoid robots can successfully perform complex, safety-critical surgical tasks (cholecystectomy) using standard manual instruments, achieving a level of instrument fidelity and control previously unobserved in early-stage research platforms.
2. Gap Analysis: While feasible, a significant gap remains between current humanoids and established surgical robots. Key limitations include control latency, reduced workspace reachability due to compact form factors, sensitivity to calibration drift, and the need for frequent system recalibration.
3. Clinical Readiness: The system is not yet clinically ready for human deployment. The authors emphasize that issues regarding sterility (lack of autoclavable components), range of motion, and workflow interruptions must be resolved.
4. Future Potential: If these technical and workflow challenges are addressed, humanoid platforms could offer a versatile, lower-cost alternative to specialized surgical robots, potentially improving accessibility and reducing operating room crowding by integrating seamlessly into existing human-centric surgical infrastructures.

The study concludes that while the "superhuman surgery" vision is not yet realized, the path forward involves iterative engineering improvements in control accuracy, robustness under dynamic conditions, and system integration to bridge the gap between research prototypes and clinical reality.