Test case design for context-aware autonomous vehicles: an in-silico experiment to observe the effects of unknown-knowns in test cases

The automotive industry is evolving towards the increased use of automatic driving aids with a view towards full self-driving automation. This brings an increased importance of software (and its quality) within the automotive. As a result, autonomous self-driving vehicles exhibit the properties of self-adaptive context-aware software systems (CASS). This paper bridges the state-of-the-art practices of testing CASS, with a view towards identifying the hazards that these can create for safety-critical systems like self-driving vehicles. To do so, we coded an in-silico experiment taking advantage of a deep learning-reinforcement learning algorithm available in stock through OpenAI Playground and Stable Baselines. Through this in-silico experiment, we highlight the hazards of the current state of the art and propose a framework to guide the research and practice of testing CASS automotive systems.