Anatomical Mesh-Based Virtual Fixtures for Surgical Robots*

Zhaoshuo Li, Alex Gordon, Thomas Looi, James M. Drake, Christopher R. Forrest, Russell H. Taylor

Abstract

This paper presents a high-performance system for generating virtual fixtures directly from complex 3D anatomical meshes. By employing a Principle Direction Tree (PD-Tree) for rapid spatial queries and a novel three-condition algorithm to handle local geometry, the system enables surgical robots to navigate intricate structures at haptic-rate speeds. Experimental results on pediatric skull meshes demonstrate significant improvements in cutting accuracy, safety, and path smoothness.

Introduction

In delicate procedures such as craniosynostosis surgery, surgeons must navigate tools with extreme precision to avoid damaging underlying soft tissues like the dura mater. Robotic assistants can enhance safety through Virtual Fixtures (VF)—software-defined constraints that restrict tool motion to a safe workspace. However, traditional VF methods often rely on simplified geometric primitives or smooth approximations, which fail to capture the intricate details of real anatomy. Furthermore, many mesh-based approaches struggle with non-convex regions, leading to system "lock-ups." This work introduces a robust, mesh-independent framework for generating dynamic guardrails for arbitrary 3D shapes.

Methods

The proposed system translates a 3D polygon mesh into a set of linear inequalities for a quadratic programming (QP) optimizer. The framework consists of several key technical innovations:

Motion Volume & PD-Tree

To maintain haptic-rate performance, the system restricts geometric checks to a "motion sphere" representing the maximum possible robot displacement per step. A Principle Direction Tree (PD-Tree), a specialized KD-Tree variant, is utilized to identify relevant triangles within this volume in microseconds.

Three-Condition Algorithm

To address the challenges of complex geometry, the algorithm classifies the tool's relationship to the nearest triangle into three distinct cases:

  1. In-triangle: The closest point lies on the triangle face, and a standard plane constraint is applied.
  2. On-edge (Convex): When the tool is near a convex edge, the edge is approximated as a plane to facilitate smooth sliding.
  3. On-edge (Concave): In concave regions, the system utilizes face normals to prevent wall penetration and avoid overlapping forbidden regions.
Closest point (CP) calculation for in-triangle (a) and on-edge (b) cases.

System Integration

The algorithm was integrated into the da Vinci Research Kit (dVRK) and paired with a piezoelectric bone cutter. This setup allows for high-frequency bone cutting while minimizing risk to adjacent soft tissues.

Enumeration of a locally convex shape to handle complex geometry and avoid forbidden region overlaps.

Results

The system demonstrated exceptional computational efficiency and surgical efficacy:

  • Performance: The algorithm achieved a refresh rate of 180 Hz for a mesh with 1 million triangles and exceeded 1 kHz for meshes under 200,000 triangles, meeting the gold standard for smooth haptic feedback.
  • Accuracy: In skull-cutting experiments, VF assistance reduced the Cutting Path Deviation (CPD) from 0.63mm to 0.37mm.
  • Safety: Mean absolute error in penetration depth was reduced from 0.83mm to 0.39mm, significantly enhancing protection for underlying tissues.
  • Smoothness: Path analysis using Spectral Arc Length (SAL) confirmed that VF-assisted motions were statistically smoother (p-value = 1.9E-4), reducing jerky manual corrections.

Conclusion

This research proves that raw, high-resolution anatomical meshes can be used directly for virtual fixtures without simplification. While the current model assumes rigid, static anatomy, future work will focus on incorporating real-time tissue deformation and improving registration accuracy between the virtual model and the physical patient. Clinical validation with multiple surgeons is also planned to further assess the system's utility in realistic surgical workflows.

Summary updated

Abstract

This paper presents a dynamic constraint formulation to provide protective virtual fixtures of 3D anatomical structures from polygon mesh representations. The proposed approach can anisotropically limit the tool motion of surgical robots without any assumption of the local anatomical shape close to the tool. Using a bounded search strategy and Principle Directed tree, the proposed system can run efficiently at 180 Hz for a mesh object containing 989,376 triangles and 493,460 vertices. The proposed algorithm has been validated in both simulation and skull cutting experiments. The skull cutting experiment setup uses a novel piezoelectric bone cutting tool designed for the da Vinci research kit. The result shows that the virtual fixture assisted teleoperation has statistically significant improvements in the cutting path accuracy and penetration depth control. The code has been made publicly available at https://github.com/mli0603/PolygonMeshVirtualFixture.
BibTeX
@inproceedings{li2020anatomical,
  author = {Zhaoshuo Li and Alex Gordon and Thomas Looi and James M. Drake and Christopher R. Forrest and Russell H. Taylor},
  title = {{Anatomical Mesh-Based Virtual Fixtures for Surgical Robots*}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2020},
  url = {https://openreview.net/forum?id=SsMIvoj_ds_},
}
Figure