DopRIO: Doppler strengthened tightly-coupled 4D millimeter-wave radar–inertial odometry

Millimeter-wave radar has long served as a key sensing modality in robotic navigation and autonomous driving due to its capability to function effectively under diverse weather conditions. The increasingly adopted 4D millimeter-wave radar further allows for the possibility of using radar to achieve all-weather odometry for robots and vehicles, which could be used to compensate for the vulnerability of widely used camera/LiDAR-based odometry methods in adverse weather scenarios. However, the inherent measurement sparsity and high noise of 4D millimeter-wave radar impose significant limitations on the accuracy of odometry estimation. This paper presents DopRIO, a Doppler-strengthened, tightly-coupled odometry method that integrates 4D millimeter-wave radar and inertial measurements and mitigates the detrimental effects of sparsity and noise by taking full advantage of the high-precision Doppler velocity measurement from 4D millimeter-wave radar. The Doppler velocity measurement is leveraged in three aspects: (1) to refine the radar points to reduce the challenges of radar point cloud registration from the source, (2) to further improve point cloud registration by considering radar point uncertainty estimated during the point refinement, (3) to improve the accuracy of state estimation by considering radar point uncertainty and utilizing a joint residual of registration and Doppler in state filtering. Experimental results on physical experimentation and public datasets demonstrate that our method outperforms existing 4D radar-inertial odometry methods and can even achieve performance comparable to the state-of-the-art camera/LiDAR-inertial odometry method. We have open-sourced DopRIO on Github¹ to facilitate the exploration of 4D millimeter-wave radar odometry.