STITCH 2.0: Surgical Robotics and Solar MHD

• The paper introduces a modular pipeline for autonomous robotic suturing on the da Vinci Research Kit, achieving sub-millimeter needle pose accuracy and significantly improved wound closure rates.
• It employs advanced segmentation and 3D suture alignment using U-Net, SAM, and RANSAC-based methods to reliably extract suture landmarks from stereo images.
• Additionally, the framework outlines a conceptual model for solar MHD that leverages data-driven subgrid-scale helicity injection to enable realistic coronal eruption predictions.

STITCH 2.0 designates two distinct research frameworks in the technical literature: (1) an enhanced robotic suturing pipeline in surgical robotics and (2) a hypothetical next-generation subgrid-scale model for helicity condensation in solar coronal magnetohydrodynamics (MHD). The following article details both frameworks with primary emphasis on their architectures, algorithmic innovations, empirical performance, and directions for future research.

1. STITCH 2.0 in Robotic Suturing: Framework and Architecture

STITCH 2.0 extends the "augmented dexterity" paradigm for autonomous robotic suturing on the da Vinci Research Kit (dVRK), achieving significant advances in wound closure accuracy, needle localization, and thread management. The pipeline is modular, with three sequential operational phases:

• Phase A: Preprocessing & 3D Suture Alignment
  1. Stereo Image Segmentation using U-Net combined with the Segment Anything Model (SAM).
  2. Flood-Fill and Skeletonization for specular noise suppression and suture landmark extraction.
  3. Fully automated 3D Suture Alignment for entry and exit point determination.

• Phase B: 6D Needle Pose Estimation with Extended Kalman Filter (EKF)

  1. Stereo Depth Reconstruction using RAFT-Stereo.
  2. Needle Point-Cloud Filtering employing Flood-Fill and Skeletonized U-Net masks.
  3. RANSAC-Based Circle Fitting for coarse pose recovery.
  4. EKF-Based Pose Refinement and tip localization.

• Phase C: Augmented Dexterity Control

  • Motion primitives include needle insertion, thread sweeping, extraction and cinching, pre-insertion alignment, and bimanual needle handover.

Tight sequential coupling between modules ensures that each phase supplies resolved and denoised state estimates to the next, thereby minimizing cumulative error—vital for high-precision autonomous operation (Hari et al., 29 Oct 2025).

2. Algorithmic Innovations

EKF-Based Needle Pose Refinement

To address unstable needle localization, STITCH 2.0 introduces an EKF instantiated on a 13-dimensional state vector: $$\mathbf{x} = [ c_x, c_y, c_z,\ l_x, l_y, l_z,\ r_x, r_y, r_z,\ n_x, n_y, n_z,\ r ]^T$$ where components encode the circle center, endpoints, plane normal, and needle radius. The dynamics are modeled as stationary ($\mathbf{F} = \mathbf{I}_{13}$), with per-trial measurement ($\mathbf{H}$) and empirical process/measurement covariances. Observations originate from RANSAC-fitted circle parameters derived from filtered point clouds. Needle tip localization involves projecting image-plane skeletons into 3D space and intersecting with the EKF’s refined circle (Hari et al., 29 Oct 2025).

Automated 3D Suture Alignment

STITCH 2.0 eliminates manual suture positioning through an autonomous, geometric pipeline:

1. Scene point cloud $P_{\rm scene}$ is generated via RAFT-Stereo.
2. Segmentation distinguishes $P_{\rm wound}$ and $P_{\rm phantom}$ based on SAM masks.
3. RANSAC plane fitting calculates wound and base planes, then computes wound height $h = |d_p - d_w|$.
4. Suture entry/exit points are distributed along a fitted line $L(t)$, ensuring uniformity and correct tissue penetration geometry.

This method minimizes point-to-surface distances and reliably aligns sutures orthogonal to the wound (Hari et al., 29 Oct 2025).

Coordinated Thread Management

To mitigate prevalent failure modes from thread cross-stitching, a "thread sweeping" subroutine actively maintains thread positioning throughout extraction and cinching. This routine, executed via precise gripper trajectories, suppresses thread entanglement by keeping the thread ahead of the needle as it emerges, as documented in Fig. 5 of the reference (Hari et al., 29 Oct 2025):

$\mathbf{F} = \mathbf{I}_{13}$5

3. Quantitative Performance and Comparative Evaluation

Empirical validation on a standard phantom using the dVRK system demonstrates marked improvements. Table 1 below provides comparative metrics between STITCH 1.0 and 2.0, as reported (Hari et al., 29 Oct 2025):

Metric	STITCH 1.0	STITCH 2.0
Avg. sutures/trial	2.93 ± 0.70	4.87 ± 0.83
Single-suture success (%)	69.4	86.9
Wound closure rate (%)	22.3	74.4
Time per suture (s)	159.3	98.74
Needle pose success (%)	42	97.56

Ablation studies confirm significant individual benefits from both EKF needle pose estimation and improved thread management—most notably a 66% increase in successful sutures, 38% reduction in suturing time, and over 3× improvement in wound closure rate. Under "augmented dexterity" (≤2 human interventions/trial), the system attains 100% closure across all runs (Hari et al., 29 Oct 2025).

4. Key Improvements Over STITCH 1.0

STITCH 2.0 enumerates seven enhancements addressing the principal limitations of its predecessor:

1. EKF-based needle pose refinement yields sub-millimeter endpoint accuracy (down from 3–5 mm).
2. Adaptive depth thresholding for robust needle tip localization.
3. Automated, point-to-surface-minimizing 3D suture alignment.
4. Coordinated thread management significantly deters thread tangling.
5. Needle-to-wound alignment module ensures orthogonal approach and endpoint leveling.
6. Flood-filled stereo imaging to eliminate specular artifacts impacting depth estimation.
7. Skeletonized U-Net masking for accelerated and robust circle fitting (Hari et al., 29 Oct 2025).

These innovations collectively enable more than a doubling of suture count, a tripling of wound closure efficacy, and substantial reductions in per-suture cycle time.

5. Limitations and Future Directions

Current constraints include non-negligible insertion height errors originating from dVRK kinematic drift (sub-mm scale), a bottleneck at 1 Hz in EKF-based needle pose estimation, and restrictive assumptions regarding wound geometry (raised, linear) and stereo camera configuration. This suggests generalization to more complex, endoscopic, or nonlinear wound architectures remains outstanding.

Proposed research directions include RNN-based calibration to correct sub-millimeter kinematic offsets, accelerated needle tracking via cropped stereo or monocular keypoint fusion, and extending the system to arbitrary wound topologies and narrow-baseline vision systems (Hari et al., 29 Oct 2025).

6. STITCH 2.0 in Solar Magnetohydrodynamics: Conceptual Prospects

Distinct from the surgical robotics context, STITCH 2.0 also refers to a hypothetical extension of the "STatistical InjecTion of Condensed Helicity" (STITCH) model in coronal MHD simulation (Dahlin et al., 2021). The original STITCH introduces a subgrid-scale electric field term,

$E_c = -\,\hat{n}\times\nabla_s(\zeta\,B_n)$

where $B_n$ is the normal magnetic field at the lower boundary, and $\mathbf{F} = \mathbf{I}_{13}$0 encapsulates the statistical effects of photospheric vortical flow. This term injects shear and relative helicity into the corona efficiently without resolving individual convective cells.

Suggestions for a future STITCH 2.0 in this domain include:

• Real-time, data-driven $\mathbf{F} = \mathbf{I}_{13}$1 from observational helicity-flux maps.
• Multi-layer injection schemes to mimic vertical helicity transport.
• A spectrum of $\mathbf{F} = \mathbf{I}_{13}$2 for scale-dependent injection consistent with a turbulent cascade.
• Direct coupling to coronal heating via an added energy equation term.
• Adaptive, $\mathbf{F} = \mathbf{I}_{13}$3-dependent smoothing to minimize over-shearing in regions with sharp $\mathbf{F} = \mathbf{I}_{13}$4 gradients.
• Rigorous stability and convergence studies for calibration against fully resolved simulations.

The conceptual STITCH 2.0 would thereby extend both flexibility and realism, enabling self-consistent, operational, and data-driven coronal eruptive event modeling (Dahlin et al., 2021).

7. Significance and Research Impact

STITCH 2.0 in robotic suturing establishes new state-of-the-art benchmarks in autonomous wound closure by tightly integrating perception, estimation, and dexterous control, with practical implications for reducing inter-operator variability and enhancing surgical consistency. Its modular, validated pipeline significantly advances the prospects for safe and effective autonomous or semi-autonomous surgical assistance (Hari et al., 29 Oct 2025).

Meanwhile, the hypothetical STITCH 2.0 in solar MHD underscores the value of simplified, data-driven, subgrid-scale models for fast, realistic coronal eruption prediction. The broader implication for both domains is that well-conceived modular upgrades leveraging recent advances in estimation, machine learning, and geometric reasoning can produce system-level performance gains unachievable by local refinements alone (Hari et al., 29 Oct 2025, Dahlin et al., 2021).