TL;DR: Augmented suturing with the da Vinci robot

Augmented Dexterity

Given the high-risk nature of surgery, there are safety concerns about having fully autonomous robots perform surgical tasks like suturing. In contrast, Augmented Dexterity allows for robots handle specific subtasks autonomously, but with a human surgeon watching closely and ready to step in at a moment's notice.

STITCH 2.0 embodies this approach. While it's designed as an autonomous suturing pipeline, it allows human surgeons to intervene and provide guidance on the fly, without having to stop and restart the entire system.

STITCH 1.0 vs STITCH 2.0

STITCH 2.0 outperforms STITCH 1.0 by completing more sutures in less time.

Additionally, STITCH 2.0 delivers higher quality sutures that are evenly spaced and close the wound.

How it works

3D Suture Alignment

We determine the suture entry and exit points by segmenting the wound pointcloud and evenly distributing sutures along the wound.

Needle Pose Estimation

We estimate the needle pose by segmenting the needle pointcloud, fitting a 3D circle to it, and refining the tip position with an Extended Kalman Filter.

Augmented Dexterity Control

Based on the suture alignment and needle pose estimation, we can perform augmented suturing with 5 subtasks: needle insertion, thread slack management, needle extraction, needle handover, and needle alignment.

Limitations

STITCH 2.0 sometimes fails in needle insertion and handover alignment because of millimeter-level positioning errors stemming from the dVRK's cable-driven design. We performed a deep calibration to reduce the error under 1 mm, but even that level of precision isn't always enough.

@article{hari2025stitch,
  title={STITCH 2.0: Extending Augmented Suturing with EKF Needle Estimation and Thread Management},
  author={Hari, Kush and Chen, Ziyang and Kim, Hansoul and Goldberg, Ken},
  journal={IEEE Robotics and Automation Letters},
  year={2025},
  publisher={IEEE}
}