RadarTrack: Radar-Only Odometry Beyond Doppler Limits

Accurate ego-motion estimation at extremely slow speeds is crucial in robotic surgery, micro-robotics, wearable sensing, and structural health monitoring (SHM), where motion occurs at millimeter or micrometer scales. Existing IMU, vision, LiDAR, and doppler-based radar methods struggle due to drift, low micro-acceleration sensitivity, or reliance on favorable environments. We propose RadarTrack , a lightweight, phase-based odometry framework that uses a single-chip mmWave FMCW radar to recover ego-velocity and full 2D trajectory without auxiliary sensors or deep learning. RadarTrack introduces a kinematics-driven phase model for sub-doppler speed estimation and a lightweight ICP-based scan association module for rotation recovery, forming a complete radar-only pipeline suitable for embedded platforms. We evaluate RadarTrack across UGV, UAV, and handheld platforms under diverse indoor conditions. Experimentally, RadarTrack demonstrates low-centimeter-per-second velocity estimation, with measurements evaluated down to 1.5 cm/s, and achieves less than 3% trajectory drift in the evaluated UGV trajectories. The system operates in real time with 0.29 s latency on a Jetson Nano, RadarTrack reduces computation by \(85\% \) compared to the closest baseline that requires approximately 3 s. As a proof-of-concept, we further demonstrate phase-based displacement tracking on a cantilever specimen under controlled loading, illustrating the potential of radar-based sensing for non-contact structural monitoring.