A Methodology for Autonomous Navigation and Mapping in an Unknown Unstructured Dynamic Indoor Environment
Abstract
Unmanned aerial vehicles (UAVs) became an effective technology for indoor search and rescue operations, providing real-time mapping of the environment, locating victims, and determining the hard-hit areas after a natural disaster. Typically, most of the indoor missions’ environments could be unknown, unstructured, and/or dynamic. Therefore, navigation of UAVs in such environments is addressed by Simultaneous Localization and Mapping approach (SLAM) in either local or global scan matching approaches. SLAM approaches that utilize laser rangefinders depend on a scan matching method of the successive scans. The local approaches suffer from high time consumption due to iterative fashion of the scan matching method. Moreover, point-to-point scan matching is prone to bad data association process. Thus, a preceding initialization step is proposed before the local approach. This step aims to increase the convergence probability and to decrease the time consumption by limiting the number of iterations needed to reach convergence. However, the local approach still suffers from accumulated errors.
Hector SLAM algorithm, as a global approach, suffers from getting trapped in local minima because of the employed gradient ascent. Hence, the multi-resolution map representation is utilized to avoid getting trapped in local minima. However, this approach increases the time consumption and the memory requirements of the process. Thus, a preceding initialization step is proposed before the Hector SLAM algorithm. This step aims to reduce the process time consumption and decrease the multi-resolution map representation into a single level with small grid cell size. However, the scan matching process of the Hector SLAM algorithm still suffers from accumulated errors.
Therefore, a low-cost novel method for 2D real-time laser scan matching based on reference key frame is proposed. The proposed method is a hybrid scan matching technique comprised of feature-to-feature and point-to-point approaches, using single laser scan rangefinder, and optical flow sensors. Unlike the local and global approaches, the proposed algorithm aims at mitigating errors accumulation using the key frame technique, which is inspired from video streaming broadcast process.
The SLAM approach is implemented using a UAV. In this scenario, the UAV can translate and rotate around all its axes. Consequently, navigating in 3D environments often requires 3D representation of the environments which usually suffers from memory and computational costs. Thus, an efficient 3D SLAM approach is proposed using multiple 2D point cloud slices.
Furthermore, for autonomous exploration, the UAV should be able to mimic humans and take decisions according to the surrounding situations. Hence, the vehicle must be able to detect a proper destination and generate an appropriate path to that destination. Since the time constraint is a key factor for most indoor search and rescue operations, an efficient exploration algorithm is proposed to maximize the visited area and minimizing the risk on the generated path.
In conclusion, to validate and evaluate the proposed algorithm, the mapping performance and time consumption of the proposed algorithm are compared with the Hector SLAM, ICP, and feature-to-feature registration such as corners, in static and dynamic environments. The performance of the proposed algorithm exhibits promising navigational and mapping results and very short computational time; the transformation parameters between each two successive scans are estimated in approximately 9 milliseconds, that indicates the potential use of the new proposed algorithm with real-time systems.
Description
Keywords
Engineering--Electronics and Electrical, Robotics
Citation
Mohamed, H. A. (2017). A Methodology for Autonomous Navigation and Mapping in an Unknown Unstructured Dynamic Indoor Environment (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca. doi:10.11575/PRISM/24788