Ivan A. Fernandez and Jingdao Chen have built a simulated testbed for stress-testing the perception and tracking software of autonomous counter-drone systems, according to a short paper on arXiv. In NVIDIA Isaac Sim, a simple quadrotor flies scripted paths while virtual sensors record synchronised camera, lidar and infrared-like data.
The testbed varies the target's range and motion, clutter, occlusion and which sensors work. Planned cases include sensor dropouts, timing errors, disagreement between sensors and false observations. In a first example, the authors tracked how a tracking filter's position error and uncertainty grew as lidar data degraded and then disappeared during a 20-second manoeuvre.



