Real-Time Path-Planning using Depth/INS Sensor Fusion for Localization

Abstract

This thesis discusses a hardware implementation of a navigating robot that plans its path and localizes itself in indoor and outdoor environments. It aims to enhance the process of navigation by enhancing the localization module of it. After feeding the cost-map along with the obstacles, path-planning algorithm (A*) chooses the lowest cost until it reaches the destination. The path planning controller uses the dynamic state of the robot using different localization techniques for outdoor and indoor environments. The objective is to make the localization process more stable to improve the navigation. The enhanced localization technique uses the depth camera localization data in indoor environments and outdoor environments, to enhance the results obtained by the IMU raw data through a depth-inertial fusion. The fusion algorithm is based on feed-forward cascade correlation network, which is part of the neural networks and it is assessed using different sensors and different neural networks methods. The robot uses the localization data in addition to the planned path to control its movement towards the destination. Results section shows the enhancement that was made to the localization process compared to the IMU localization or the depth camera localization through using the mentioned fusion algorithm.