LAUSR

Radar has emerged as a core sensing technology for many active-safety and comfort related advanced driver assistance systems (ADAS) being deployed in today’s vehicles. Using radar technology, the ego vehicle can simultaneously detect the range and velocity of multiple targets. With multiple-input, multiple-output (MIMO) arrays, it is also possible to detect the angle of arrival of targets in azimuth and elevation. 4D automotive radar sensors can determine the range, velocity, azimuth and elevation angles-of-arrival (AoA) of targets in a traffic scene. Currently, radar-returns from traffic scenes have been mainly obtained through measurement. While measurement is valuable, it can be costly, time consuming, restrictive due to practicality issues and also unsafe in some corner-case traffic scenarios. Simulation has emerged as a potential alternative source of synthetic radar-returns that can be used to develop, test and refine signal processing techniques and detection algorithms. A key challenge in simulation has been to retain an accurate representation of actors in full-scale traffic scenes while still being able to solve electrically large problems efficiently. The introduction of MIMO arrays further increases the complexity and computational load demands of such simulations. In this paper, we present a computationally efficient, high fidelity, physics-based simulation workflow for a 77 GHz frequency-modulated continuous-waveform (FMCW)-based, 512-channel MIMO radar sensor. We demonstrate how the synthetic radar returns obtained from full-scale traffic scene simulations can be used to create 4D-radar point clouds. The accuracy of the synthetic radar returns is then evaluated by overlaying the resulting 4D-radar point clouds on 4 corresponding full-scale traffic scenes with varying levels of complexity. Results from this study demonstrate how accurate radar returns obtained from simulation can be used to develop next-generation radar sensors for autonomous vehicles.