A major downside of simulation environments for autonomous systems (e.g. CARLA) is the lack of RADAR simulation. While it is super easy to simulate realistic LiDAR and camera data for most situations (excluding a few very special cases here), it is very challenging to simulate RADAR signals. The perhaps greatest challenge is to estimate the reflective properties of the environment and other vehicles/pedestrians/etc. and their change with frequency shifts. Let’s have a brief look at this problem.
RADAR offers unique capabilities in dirty environments as well as in “challenging weather conditions” (heavy rain and snow fall). Further, it allows tracking up to 2 (or 3?) cars using RADAR back scatter in 2016 (that might be invisible to LiDAR or cameras. Delphi claims to provide a single RADAR solution to identify 64 targets in a vehicle’s path.
It seems like LGSVL offers some kind of RADAR simulation capabilities. I’m not exactly sure how it is implemented but I assume that they use ray tracing and don’t solve Maxwell’s equations on a FEM grid. What I’m looking for is a RADAR simulation that allows for different waveforms, especially the new stuff that manufacturers use in R&D only because it would require more computing power - welcome to the 1980s ;).