Reducing Drift in GNSS-Denied Environments with Radar Odometry

In this paper, a novel radar odometry framework called DeRO is proposed, which computes poses by directly using the Doppler velocity for dead reckoning of a 4D FMCW radar combined with gyroscope data. The error accumulation caused by accelerometer bias and double integration that are present in conventional approaches are avoided with this method. To increase accuracy, the system estimates a radar velocity scale factor and updates measurements in a Kalman filter using accelerometer-measured tilt angles and radar scan matching. In comparison to the most advanced radar-inertial odometry, the approach lowers average position error by 47% and rotation error by 52% when tested on real-world datasets.