Novel Pose Estimation System for Precise Robotic Manipulation in Unstructured Environment

摘要

Intelligent robotic systems are becoming essential for industry and harsh environments, such as the CERN accelerator complex. Aiming to increase safety and machine availability, robots can help perform repetitive and dangerous tasks, which humans either prefer to avoid or are unable to do because of hazards, size constraints, or the extreme environments in which they take place, such as outer space or radioactive experimental areas. A fundamental part of intelligent robots is the perception of the environment that is possible to obtain only knowing the 6D pose of the objects around the robotic system. In this paper, we present a novel algorithm to estimate the 6D pose of an object that can be manipulated by a robot. The proposed algorithms works consistently in unstructured and harsh environments presenting several constraints like variable luminosity, difficult accessibility and light reflections. The algorithm detects the position and rotation of an object using 3D cameras. The procedure has been developed using Point Cloud Library to manage the point cloud created with an RGBD Camera. The position and rotation of an object is useful in augmented reality systems to help the tele-operator and for the realization of autonomous or semi-autonomous tasks.