Automating Multiple-Throw Multilateral Surgical Suturing with a Mechanical Needle Guide and Sequential Convex Optimization

Suturing is a ubiquitous sub-task in surgery. To explore supervised automation of multi-throw suturing in Robot-Assisted Minimally Invasive Surgery, we present a novel mechanical needle guide design and an optimization framework to optimize needle size, needle trajectory and control parameters for two arms using sequential convex programming. We also develop a model-based needle tracking system to achieve closed loop control. We show that in comparison with the standard 8mm needle driver, our Jaw-mounted Needle Guide (JNG) improves accuracy of needle orientation by 3x on average in the presence of needle orientation noise of up to 30◦ in either axis, and reduces the need for needle realignment before subsequent needle insertion. Use of real-time needle tracking can estimate pose with a standard deviation of 2.94 mm and 6.8◦. We demonstrate our framework on a 4throw suturing task on a da Vinci Research Kit using tissue phantoms for skin and subcutaneous fat. We evaluate suturing performance based on success rate and compare completion time with suturing demonstrations in JIGSAWS dataset [5]. We evaluate effects of pose constraints on the suture path with respect to path length and tissue trauma. Our results indicate that dVRK can perform 4-throw suturing at ≈ 1/3rd of human speed at a success rate of 50% for the 4-throw task, successfully completing 86% suture throws attempted. Task videos and data available at: berkeleyautomation.github.io/amts