When modeling technical systems as black-box models, it is crucial to obtain as much and as informative measurement data as possible in the shortest time while employing safety constraints. Methods for an optimized online generation of measurement data are discussed in the field of Active Learning. Safe Active Learning combines the optimization of the query strategy regarding model quality with an exploration scheme in order to maintain userdefined safety constraints. In this paper, the authors apply an approach for Safe Active Learning based on Gaussian process models (GP models) to the high pressure fuel supply system of a gasoline engine. For this purpose, several enhancements of the algorithm are necessary. An online optimization of the GP models' hyperparameters is implemented, where special measures are taken to avoid a safety-relevant overestimation. A proper risk function is chosen and the trajectory to the sample points is taken into account regarding the estimation of the samples feasibility. The algorithm is evaluated in simulation and at a test vehicle.

No abstract

No abstract

In order to handle the challenges of autonomous driving, deep learning has proven to be crucial in tackling increasingly complex tasks, such as 3D detection or instance segmentation. State-of-the-art approaches for image-based detection tasks tackle this complexity by operating in a cascaded fashion: they first extract a 2D bounding box based on which additional attributes, e.g. instance masks, are inferred. While these methods perform well, a key challenge remains the lack of accurate and cheap annotations for the growing variety of tasks. Synthetic data presents a promising solution but, despite the effort in domain adaptation research, the gap between synthetic and real data remains an open problem. In this work, we propose a weakly supervised domain adaptation setting which exploits the structure of cascaded detection tasks. In particular, we learn to infer the attributes solely from the source domain while leveraging 2D bounding boxes as weak labels in both domains to explain the domain shift. We further encourage domain-invariant features through class-wise feature alignment using ground-truth class information, which is not available in the unsupervised setting. As our experiments demonstrate, the approach is competitive with fully supervised settings while outperforming unsupervised adaptation approaches by a large margin.