Systematic Evaluation of Autonomous Parking Modules Using Constrained Hardware

Abstract

In modern vehicles, autonomous features are highly sought after. However, the complexity of the problem domain and the security aspects make the software expensive to develop. Therefore, autonomous features are available only in the most exclusive vehicles. If proper software engineering procedures can be applied in the domain of autonomous driving, the cost of development can be reduced, increasing the number of autonomous features on the market. This thesis aims to research how systematic evaluation can be performed on a series of software modules. The modules were developed as a proof of concept that it is possible to achieve specific AD features using fewer sensors and cheaper hardware than what currently is used in the automotive industry. The development and testing are done working in an agile manner, following the design science framework. All the separate software modules are systematically evaluated individually to confirm their viability and compared to similar algorithms to decide which ones would be the best fit for the project. This is done mainly through experimental studies in a controlled environment. The results of the systematic evaluation show that the implemented methods for parking space detection, pathfinding, path following, and parking algorithm are viable and can accurately detect and locate a parking space. The pathfinding algorithm can then generate a valid path to a desired starting position which the path follower can use to relocate the platform to that position with adequate accuracy. However, the choice of using BreezySLAM due to the constrained hardware proved to be a problem because of a large amount of spatial drift in the coordinates. Even though the specific testing methods used in this thesis are tailored to test the specific types of algorithms used in the project, the general methodology of how the tests were designed and the practice of utilising systematic evaluation throughout the development of an artifact could potentially be generalised and utilised in any, or most software engineering project.