Multi-section cable-driven continuum surgical manipulator by learning from demonstrations

Abstract

In cable-driven continuum manipulators, the cables are used to transmit the movement of the actuators to the joints. Therefore, the actuation system can be located remotely, which offers enormous possibilities for robot assisted minimally invasive surgeries (MIS). MIS is a form of surgery intended to provide great benefits to patients over conventional open surgery by minimizing unnecessary trauma caused in the process of performing a medical procedure. MIS leads to less pain, blood loss and better cosmesis, and reduces the recovery time for patients. However, due to the highly limited workspace, indirect visual and tactile information feedback, and specialized tools, MIS is very hard for surgeons to perform, which highlights the importance of robot assisted MIS. In this paper, we suggest to use a novel motion planning paradigm, learning from demonstrations (LfD), to transfer three benchmarking clinical skills, e.g., insertion, targeting, and obstacle avoidance, for a cable-driven continuum manipulator. The advantages of the proposed approaches mainly lie on four folds: Firstly, a cable-driven manipulator with two sections (4 degrees-of-freedom) is implemented. Secondly, based on the piecewise constant curvature assumption, a parameterized model is derived to describe the kinematics of the cable-driven manipulator. Thirdly, an expectation-maximization based policy search algorithm is implemented to iteratively update the derived parameterized model to compensate for the nonlinearities in the system. And finally, a human operator teleoperates the manipulator to accomplish the aforementioned three tasks several times, then Gaussian Mixture Models and Gaussian Mixture Regression are applied to encode the demonstrations with a dynamical systems model and generalize corresponding actuator movements for the manipulator to reproduce the learned skills. Our newly proposed cable-driven continuum manipulator is a soft snake-arm robot. Compared with the structures of formerly developed soft robots, this original prototype has three major superiorities. Firstly, each vertebra is designed as a flake-shaped node and the backbone is made by a 0.8-mm diameter Austenite-stainless-steel wire. This eliminates the friction between the adjacent nodes and reduces total weight of the skeleton to less than 10g, thus highly increasing the response rate and allowing the arm to reach a maximum bending angle of 17.5 degree for each joint. Secondly, the appropriate decrease of the reference circle diameter in the 2nd section minimizes its bending interferences on the 1st section, which improves the flexibility and controllability when both sections bend. Thirdly, two groups of steel cables (two pairs in each group for one section) are introduced to actuate the manipulator through via holes. The collaboration of the two groups of cables enables the manipulator to bend in an S-shape.