A Deep-Learning Model of Virtual Test Drivers

Virtual test drivers are becoming a paramount automatic verification tool enabling car makers to test new and advanced vehicle functionalities in a standardized, repeatable, high-quality, and cost-saving way. In this letter, we use modern machine-learning methodologies to build a virtual driver able to test the hill-descent control, one of the driver assistance systems equipping modern cars. The experimental results show that our virtual driver performs as a human driver involved in the same test conditions.