3-dimensional object perception for manipulation tasks using the atlas robot

Abstract

The goal of the DARPA Robotics Challenge (DRC) is the development of ground robots capable of executing complex tasks in disaster relief environments. The Virtual Robotics Challenge (VRC) was the first phase of the competition, in which teams develop their own software to maneuver a simulated robot through a virtual obstacle course and perform a set of complex tasks. In these scenarios, grasping and manipulation tasks are required; therefore this paper provides a description of the visual 3-Dimensional (3D) perception functionality to grasp a hose lying on a table, one of the tasks of the challenge. The sensor head of the used Atlas robot is equipped with a laser range scanner and a stereo camera, which provide the 3D point cloud to be processed. The presented approach processes the point cloud; a plane detector is used to segment the scene and a tabletop assumption is then considered to detect the objects of interest. The hose connector is recognized using color-based region growing segmentation and then its pose is estimated by a cylinder model fitting. The experimental results demonstrate the efficiency of the proposed 3D perception pipeline in the scenario under study.