Robust Optimization for Online simultaneous localization and 3D mapping on Quadrotor
Localization and mapping are fundamental requirements to enable mobile robots to operate robustly in their environments. In this thesis, we have created novel simultaneous localization and mapping (SLAM) benchmark which are more robust to place recognition and sensor errors. The first part of the overall problem is to create a graph, by identifying nodes and factors between the nodes based on sensor data. Such a system is often referred to as the front-end. The second part entails finding the configuration of the nodes that best explains the factors. This step corresponds to computing the maximum-likelihood map and a system solving it is typically referred to as a back-end. Converging to the correct solution is challenging if outliers are present or the initial guess is far away from the global minimum. Therefore we investigated four robust algorithms and evaluated its robustness against outliers. with consideration a trade-off between processing time and accuracy, we have emphasized on dynamic covariance scaling (DCS).This thesis has reviewed variety methods to using each published algorithms in diff erent parts of the SLAM and ultimately we have designed a dynamic and user-friendly software package with Robot Operating System (ROS) that called KNTU-SLAM. In this benchmark, The latest algorithms in different parts of the system has aggregated together. by appropriate recommendations, efficiency of feature extraction and image descriptors algorithms were improved the results since compared that with the previous powerful benchmark like RGBD- SLAM which created at University of Freiburg. The superiority of this project has shown when implemented on a variety of mobile robots at ARAS Lab in K.N. Toosi University of Technology. Also for the first time, creating a three-dimensional map and localizaton in real time on quadrotor with using Kinect camera by wifi that transfered data to another computer for processing. it has been with a new proposed node in ROS. To display the produced map has used OctoMap and point cloud that has given required components for Autonomous aims to the user.