Tying Surgical Knots From Demonstration : Enhancing Demonstrations and Correcting Errors During Execution

Recent advances in the modeling of deformable objects such as surgical suture, rope, and hair show significant promise for improving the manipulation of such objects. An important application of these tasks lies in the area of medical robotics, where robotic surgical assistants have the potential to greatly reduce surgeon fatigue and human error by improving the accuracy, speed, and robustness of surgical tasks such as suturing. Past work has shown how open-loop suture manipulation trajectories can be found through learning from demonstrations and through motion planning. It has, however, mostly left untouched the challenges of (i) Performing feedback control to allow reliable execution in the presence of error in the initial state and perturbations that disrupt the execution of the trajectory. This is a challenging task as we are faced with a highly underactuated system. And, (ii) Improving trajectories obtained from demonstrations and sampling-based motion planners, which may contain mistakes and errors in the case of demonstration or needless motion in the case of motion planning. In this paper we propose sequential quadratic programming (SQP) formulations to tackle each of these challenges, and validate our approach experimentally. Our method enables successful executions of suture manipulation tasks in the presence of significant perturbations, and it allows us to significantly enhance demonstrated trajectories. For instance, we are able to remove partial failures during demonstrations without any need for annotation of the demonstrations.