Fraunhofer Institute for High Frequency Physics and Radar Techniques FHRSCaRL is a large-scale synthetic dataset developed at Fraunhofer FHR to support the training and validation of autonomous driving systems. Built on top of our former CARLA simulator, SCaRL provides fully synchronized data from a complete and diverse sensor suite, including RGB, semantic, instance, and depth cameras, coherent LiDAR with Doppler capability, and MIMO-FMCW-radar, across 140,000 time-aligned frames covering dynamic, multi-actor traffic scenarios.
Beyond being a dataset, SCaRL is a fully-fledged radar simulation tool. Its physically grounded radar simulator is based on ray-tracing, enabling the generation of raw radar data for arbitrary sensor configurations, including custom array geometries, MIMO front-ends, and waveform parameters. This makes SCaRL particularly well suited for rapid prototyping of new radar concepts and processing algorithms in realistic dynamic scenarios, without the need for costly measurement campaigns.
The coherent LiDAR simulation is equally physics-aware, incorporating surface normals, angle of incidence, and Doppler estimation via ray-tracing, going well beyond what standard LiDAR simulators typically offer.
SCaRL is designed to bridge the gap between simulation and real-world perception by enabling sensor fusion research, neural network pre-training, and cross-modal algorithm benchmarking, with domain adaptation as a natural pathway towards deployment on real-world data.