RAGNAROK: Radar-Aided Gravity-Normalized Alignment for Robust Open Keyframe-based Radar-Visual-Kinematic-Inertial SLAM
This study addresses state estimation drift in legged robots traversing complex terrain, caused by proprioceptive disturbances and insufficient radar yaw observability. To this end, it proposes the first SLAM system integrating radar, vision, kinematics, and inertial measurements. Methodologically, we introduce a pioneering radar-aided gravity normalization alignment framework alongside a slip/roll contact-aware legged velocity estimator with adaptive weighted fusion. Furthermore, a B-spline radar-aided proprioceptive backbone network is constructed, incorporating kinematics-aware radar factors, online extrinsic calibration, and degraded image enhancement. Extensive experiments across multiple datasets demonstrate that the proposed system achieves significantly superior robustness over existing state-of-the-art methods in harsh environments.