Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces DiffRadar, a novel radar SLAM system that utilizes a differentiable, physics-aware Gaussian field to model radar observations, addressing the limitations of traditional scan matching methods. By representing the scene with anisotropic Gaussian primitives and employing a differentiable radar forward model, DiffRadar enables the joint optimization of robot pose and scene structure directly from radar data. The system demonstrates significant improvements in trajectory accuracy and map consistency, particularly in challenging environments, achieving real-time performance at 70 FPS.
Modeling radar observations directly in the signal domain leads to a more robust and consistent radar-only SLAM, outperforming traditional methods in dynamic and feature-poor environments.
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.