Polarimetric radar for self-localization in automotive environments

As automotive vehicles advance towards higher levels of automation, the static environment is becoming increasingly more important after the focus was primarily on the perception of other road users in earlier driver-assistance systems. For safely navigating through complex driving maneuvers, the vehicle must reliably know its own position and orientation with respect to its surroundings. Therefore, characteristic elements in the environment, known as landmarks, must be detected and aligned to a map for deriving the desired position. While some research on utilizing radar for this is available, no extensive study on leveraging polarimetric radar in aself-localization context is published to date. Therefore, several aspects of polarimetry for perceiving the static environment and subsequently localizing against it are addressed in this thesis. One important prerequisite for exploiting polarimetric information in a practical application is ensuring the stability of a polarimetric response over the radar’s field of view, especially in angular and speed dimensions. Thus, a tailored method for a polarimetric calibration is developed and builds the basis for public road experiments with a newly setup experimental vehicle. A polarimetric analysis of the scattering mechanisms occurring at typical automotive landmarks provides insights into the requirements for properly representing and leveraging this information for static environment perception. Apart from an analysis of an exemplary real-world outdoor measurement, undisturbed anechoic chamber measurements enable a detailed investigation of scattering effects and their spatial allocation. Both analyses reveal that anisotropic behavior is prevalent for typical landmarks and a tailored handling in the environment representation is advisable. In addition to this newly suggested polarimetric part in the proposed gridmapping approach, a technique for incorporating low-level spectral data is developed. Both extensions are the basis for more robust landmark extraction algorithms, demonstrated through extensive real-world data recordings in public road traffic. It is shown that the polarimetric information makes the extraction of point-shaped landmarks more robust, whereas the low-level data incorporation is advantageous for reliably detecting line-shaped landmarks. Based on this preliminary work, a self-localization framework functioning solely on radar data is presented. The framework employs a factor graph formulation for an efficient implementation and is evaluated using several hours of real-world recordings in diverse environments. In addition to generally promising accuracy of the results, the influence of the polarimetric information is investigated in detail. While in environments with an abundance of landmarks, the localization accuracy doesnot improve with a higher polarimetric data level, a clear correlation between localization performance and polarimetric information content is observed in challenging environments. This proves the benefit of polarimetry for self-localization.