UC Davis

The aim of this research is to develop the capability to sense, model, and manipulate a deformable linear object in a workspace to facilitate autonomous maintenance and repair operations in a deep space habitat. The specific capability developed in this work is the ability to clear a deformable linear object that is obstructing the path to a target object, which is to be reached. The robotic system presented herein consist of two six degree of freedom arms, working in a shared workspace, along with stereo and color cameras perceiving the environment. A perception pipeline was developed to segment a wire from a point cloud and estimate the state of the wire, directly from stereo vision. The state of the wire is represented by a set of 20 nodes, equally spaced, that when connected form the shape of the wire. To determine where to move the wire, in order to clear a path to the target object, a physics-based simulator was developed to simulate different pick and pull/push actions on the wire. The pair of actions that maximizes the space around the target object were then selected to be executed by the robot. To grasp the wire, a grasp planner was developed which solves for six degree of freedom valid grasps on the wire. The proposed system was tested on hardware and demonstrated the ability to accurately sense and estimate the position of a wire in the workspace, and move the wire to clear a path to a target object. The overall system was run 30 times, for different wire configurations, and achieved a 86.7% success rate at clearing an obstructing wire.