IMPLEMENTATION OF FASTSLAM 2.0 USING VEHICLE SIMULATION ENVIRONMENT ADOPTING A FLEXIBLE ARCHITECTURE FOR SIMULATION TO VEHICLE TESTING

In a world where the development has to be incredibly fast-paced in-order to success in the market, the importance of a definition of an architecture plays an very important role in integration of the product into the target hardware and which thereby involves the extension of the product development from a simulation environment into HIL environment and the subsequent extension into the vehicle. This paperwork revolves around architecture definition for an autonomous driving car that utilizes the virtual simulation environment for development and permits the extension to processes that comes later in the development of the project and the eventual deployment. The thesis work also discusses the implementation of a SLAM technique on the defined architecture in open loop. Increasing importance of autonomous driving for transportation encompasses the need for a precise mapping of the environment, which in turn allows for an accurate and reliable localization of the ego-vehicle .This is integral to an effective decision making to integrate a reliable motion planning for autonomous vehicles. The implementation and testing of the various algorithms directly in the prototype vehicle makes the development of the algorithm painstaking and slow. This is because, analysis of the results with the groundtruth becomes difficult in a complex environment. The virtual simulation with accurate vehicle dynamics in a custom modelled environment makes the development process easier and faster. The project covers the implementation, results analysis and comparison of the particle filter based fast slam 2.0. The simulation environment used to generate scenarios is the IPG CarMaker in which several scenarios at low speeds are defined. The scenarios are developed considering the maneuvers that would be most frequent in an urban space like parking in a garage or in a tight space which would make life easier for the driver. The generated scenarios are processed in the ROS environment using designated algorithms considering the noises and reliability of odometry and perception sensors measurements. The RVIZ visualizer tool inside ROS is used for visualizing the generated maps which are occupancy grids generated from laser point clouds and the vehicle odometry. The results are analysed and the outcomes are recorded which would be used as a platform to incrementally develop a reliable and robust SLAM algorithm for urban spaces to make driving easier and autonomous.