Data for Training Models, Domain Adaptation

It is rather clear that sensor-based perception and action must be based on data-driven algorithms. Simulation could provide data to train and test algorithms that are afterwards implemented in vehicles. This section describes how virtual worlds can generate photo-realistic images with pixel-wise ground truth, which can be used to train on-board visual perception systems. Using domain adaptation, such systems can operate in real-world environments, while avoiding cumbersome manual labelling of data up to a large extent.