Radar Sculptor: How Robots Perceive a Seamless World express
A regular radar works like an amateur photographer: it takes a series of blurry snapshots and tries to stitch them into a panorama. Details are smeared, and in a long corridor or crowd, the robot easily gets lost.
DiffRadar abandons snapshots. It sculpts a living sculpture from radio signals — a fluid field that updates with each new reflection. Map and robot position are refined together, seamlessly. In bare walls, during sharp maneuvers, and among darting objects, the machine moves smoothly, as if on a thread. And all at the speed of a video camera—70 frames per second; radio waves travel at the speed of light, so measurements are almost instantaneous — even in total darkness or thick smoke.
Original abstract
as written by the authors arXiv:2607.12265
Radar sensing is increasingly used in mobile systems because it operates reliably under poor lighting, adverse weather, and privacy-sensitive settings where cameras and LiDAR often fail. However, most existing radar SLAM systems estimate motion through scan matching on discretized radar heatmaps, which breaks geometric continuity and fails to capture key radar sensing properties, often leading to unstable pose estimation and degraded mapping in regenerate or dynamically changing environments. We present DiffRadar, a real-time radar SLAM system that models radar observations as a differentiable, physics-aware Gaussian field rather than discrete scans. DiffRadar represents the scene as anisotropic Gaussian primitives and renders radar measurements in range-azimuth and Doppler-azimuth spaces through a differentiable radar forward model, enabling joint optimization of robot pose and scene structure directly from radar measurements. We implement DiffRadar on commodity FMCW radar hardware and evaluate it on both the public Radarize benchmark and a controlled stress-test suite that targets common radar SLAM failure modes, including corridor degeneracy, motion regime transitions, dynamic clutter, and long-horizon loop closures. DiffRadar achieves substantial reductions in trajectory error on the benchmark, with especially large gains under feature-poor corridor motion, while more than doubling map consistency and maintaining real-time performance at 70 FPS. These results show that modeling radar observations directly in the signal domain enables substantially more robust and consistent radar-only SLAM for mobile platforms.
FMCW police radars catch speeders with centimeter precision, and they can even detect if the driver just took a deep breath.
This vision is like that of alien ships in Banks’s novels: they sense space with every point of the hull, instantly adapting to any disturbance.