Simultaneous localization, mapping applications and object recognition based localization

Simultaneous Localization and Mapping (SLAM) problem is defined as the robot positioning itself while simultaneously creating a map of an unknown environment. Using a distance sensor such as the LIDAR (Light Detection and Ranging) sensor and an odometry sensor from which the wheel-angle values are taken, the robot estimates the map of the environment and its location. Location information in outdoor environment can be determined with the Global Positioning System (GPS) with high accuracy, while indoor environment cannot be determined due to the lack of GPS data in indoor environments. Indoor environment localization is an important issue especially for defense technologies. In a indoor environment, the robot receives sensor data as input and outputs a position estimate. It makes this prediction using various Bayes-based prediction methods. One of the most commonly used estimation methods is the Particle Filter (PF), which is not required for Gaussian Distribution. PF based Monte Carlo Localization (MCL) method is used to estimate the location of the environment. Robot odometry makes successful predictions if there is no error or irreversible noise in the sensor data. However, in the event of a problem such as wheel slippage, robot hijacking and obstruction, the robot loses angle data and makes the position estimation incorrect. In this thesis, an innovative method based on object recognition has been proposed for positioning the robot in a loss of wheel data that may occur in closed environments. The Faster R-CNN, a Deep Learning model, aims to position the robot according to the position of two recognized objects. For this purpose, experimental results are given and it is seen that the robot performs a successful position estimation in case of faulty wheel data where classical prediction methods fail. In the thesis, mapping and scaling of a closed environment without using the odometry data was performed. Gmapping and HectorSLAM methods, which are the two most commonly used methods in Simultaneous Localization and Mapping (SLAM) in the literature, have been applied and advantages and disadvantages of these methods have been indicated. Indoor positioning and indoor mapping, which is an important issue in the integration of defense technologies and defense systems, were discussed and the proposed methods proved to be successful. Keywords: Simultaneous Localization and Mapping (SLAM), Autonomous Robots, Defense Technologies and Defense Systems Integration, Particle Filter (PF), Monte Carlo Localization (MCL), Deep Learning, LIDAR (Light Detection and Ranging)