A High-precision and Robust Odometry Based on Sparse MMW Radar Data and A Large-range and Long-distance Radar Positioning Data Set

Lidar-based or vision-based positioning systems are easily affected by bad weather, and RTK-GNSS inertial navigation systems are prone to reduce positioning accuracy in environments with poor GNSS signals. And using radar for positioning can overcome the challenges of using other sensors for positioning in bad weather. However, compared with lidar, radar has more sparse data, low ranging accuracy, a lot of noise, and only two-dimensional perception results. In this paper, we propose a high-precision radar odometry method to overcome the disadvantage of sparse radar data by fusing multiple frames of radar data to form a sub-map. The error is reduced by graph optimizing the pose of the sub-map, resulting in an error of 1.737% in translation and 0.0018 deg/m in rotation. A radar-based positioning dataset was collected and organized. The comprehensive test on the dataset shows that the accuracy of the odometry is high, and the positioning frequency is higher than that of the data.