This thesis deals with state estimation schemes for horizontal vehicle dynamics.
It explores the space between classical, established approaches like Kalman Filters and approaches with learning systems, such as Neural Networks. After the definition of the mathematical concepts underlying this thesis, it introduces novel frameworks with a combination of physics-based models and learning parts. For the evaluation, this thesis lists a comprehensive testing catalog and presents the database which was collected in several vehicle testings. The application chapter shows the practical implementation of the novel approaches, derives the physics-based vehicle models, and designs application-specific Neural Networks. The results set new performance benchmarks for the specific application. The analysis of results compares estimation performance between estimators and estimation properties in a number of situations. It concludes that the models are viable approaches to complex estimation tasks. This Thesis’s title is a combination of vehicle state estimation, which describes the reconstruction of vehicle states, and the field of machine perception, which describes a computer’s capability to perceive the world and itself in a similar sense as living beings. This terminology is common in computer vision and machine learning tasks.