Autonomous Suturing

A Case Study of Trajectory Transfer Through
Non-Rigid Registration for a Simplified Suturing Scenario

John Schulman, Ankush Gupta, Sibi Venkatesan, Mallory Tayson-Frederick, Pieter Abbeel


Welcome! This website supplements our IROS 2013 submission, in which we present an algorithm for automating suturing. The algorithm generalizes demonstrated motions to adapt to previously unseen suturing environments using non-rigid registration and trajectory optimization.

Code for general geometry-based learning from demonstrations: Docs Github

The following is a video of the presentation of the paper as given at the IROS 2013 Conference in Tokyo.
It is followed by demonstrations of the proposed algorithm for various suturing settings.

Suturing with Raven II (simulation)

The algorithm was validated on a physics simulation using a Raven II surgical robot. These simulation experiments are interesting because (1) being in simulation, they are more readily reproducible and a larger number of experiments can be run, (2) begin a surgical robot, the Raven II has unique kinematics: non-redundant arm kinematics and a remote center of motion.

The Raven II robot performs a suture stitch - which means two punctures and a knot tie. To test trajectory generalization in the presence of non-rigid perturbations, the left flap of the suturing pad was kept unchanged but the right flap was perturbed by translations and rotations.

The following is a video of an experiment run on the Raven II. It includes a human demonstration and the performance of the robot for a particular perturbed setup.

Suturing with PR2

The algorithm was also tested in a mock suturing setup with the PR2 robot. The tissue was simulated using thin foam with pre-cut holes and slits. The demonstrated trajectory was manually annotated and qualitatively broken into segments such that each could be executed in open loop given the 3D data at the beginning of the trajectory. The task studied here consisted of piercing both sides of the tissue across the cut and pulling the needle out.

The following is a video of the algorithm as tested on the PR2. It includes the human demonstration and key-point annotation of the task. Also shown is a mosaic of new suturing environments to which this approach was generalized.