Vasilii Mosin presenterar sin licentiatuppsats Automotive Machine Learning-Based Perception Software Engineering - Development Phases and Data Handling.
Introduction of Machine Learning (ML), due to its probabilistic and data-centric nature, changes the way of software engineering. This also applies to the engineering of automotive perception software, which is at the core of self-driving vehicles. This thesis seeks to explore the implications of ML on automotive perception software engineering, particularly focusing on software development phases and data handling practices. Through a series of case studies and experimental studies, we aim to address this objective in collaboration with an automotive company.
Our results illustrate how ML influences automotive perception software engineering across multiple dimensions, encompassing the entire software development lifecycle, design and testing phases, data collection, data splitting, anomaly detection, and some other facets. There remain more aspects to study regarding the impact of ML, but the findings from this thesis could still help contribute to improving future automotive perception software engineering.