Human-like indoor navigation for Autonomous Industrial Mobile Manipulator

Abstract

Autonomous Industrial Mobile Manipulator (AIMM) that combines the advantages of both mobile robot and industrial manipulators and owns great mobility, flexibility and functionality will be the next generation of robots used in industrial automation. Compared to the tractional industrial robots, it is capable of performing various tasks in unstructured or semistructured environments, thus brings great challenges in autonomous localization & navigation, object identification, control and coordination. In this paper, a novel human-like indoor navigation problem is studied. Instead of predefining a feature map or building a 3D point cloud map, a human-like topological description and realtime corridor identification are utilized. As a fundamental problem, corridor classification stands for a key role in the whole system. A cascade Bayesian classifier is designed to make full use of multi-observations and gets much better confidence and more stable results. The classifier is consisted of a weak incomplete feature extractor and a strong bayesian classifier. Features are extracted from depth maps provided by a Microsoft Kinect sensor. With several observations, the Bayesian classifier fuses all the features and forms the final results. Experiments are performed on a recently built AIMM system, and the results validate the effectiveness of the proposed methodology.