Efficient pole detection and grasping for autonomous biped climbing robots

Biped climbing robots alternately use their grippers for attaching to locomotion in truss-style environments. To implement autonomous climbing, pose detection and grasping of the target pole is one of the fundamental capabilities. In this paper, we present our methodology to efficiently detect and recognize a pole for grasping based on a low-cost depth-image camera only. Each pole, the element of a truss, is parameterized to describe the structural environment and to guide the sensor data processing. Acquiring depth and image data from the camera mounting on top of the gripper, efficient algorithms are then deployed to extract, recognize and parameterize the target pole. Feasible grasping pose are finally computed considering the geometrics of both the gripper and the target pole, and the movement of each joint is obtained by solving inverse kinematics and sent for servoing to fulfill autonomous grasping. A serials of experiments have been conducted to verify that the proposed methodology and accompanying algorithms satisfy the application requirement for biped climbing robots.