How machine learning drive the development of autonomous cars - A study of emerging technologies and the evolving value chain Master’s Thesis in the Master’s Programme

Autonomous cars have gained much attention in the past few years. Partially automated features are sold to consumers, and advance autonomous test vehicles are driving on public roads. These features and vehicles have caused a debate about their safety as they have been involved in a couple of fatal crashes in recent years. To simultaneously develop autonomous cars and features without compromising safety is a big challenge for the automotive industry, which is investing heavily in technology. This thesis aims to describe constraints in advanced driver-assistance systems (ADAS) and autonomous driving (AD) developed with machine learning techniques, how companies are adapting to this new technology demand, and current development paths towards autonomous cars. A model of the development of ADAS/AD was created by interviewing industry experts and a literature review. Companies efforts in the area were identified by searching databases and reading news articles and press releases. Data is a valuable asset but requires extensive work to be useful when developing ADAS/AD. Collecting data per se is not considered as a challenging task. Instead, the difficulty is to collect data about edge-cases, that is, rare situations occurring maybe once in a human lifetime of driving. Another key takeaway is the inherent difficulty of creating consistent datasets to train neural networks with, as it requires humans to interpret subjective situations the same way. Incumbent firms in the automotive industry are accompanied by startups and established technology companies from outside the industry in trying to develop and capture value from ADAS/AD. Collaborations and acquisitions are prominent ways to get desired know-how and secure supply of critical components. A schematic overview of the relationships in the industry is mapped to give the reader an idea of the sophisticated ecosystem companies are part of through investments, acquisitions, spin-offs and collaborations. Efforts to capture value from ADAS/AD can be divided into two branches. One being OEMs developing and offering ADAS in consumer products aiming to develop autonomous driving gradually. The second is characterized by companies targeting autonomous robo-taxi solutions in geo-fenced areas. Although both approaches are facing many similar challenges, they also differ in specific areas such as operational design domain.