3DRO: Lidar-level SE(3) Direct Radar Odometry Using a 2D Imaging Radar and a Gyroscope

A research team from the University of Toronto, led by Cedric Le Gentil, has introduced 3DRO, a novel method for 3D motion estimation that rivals the accuracy of expensive lidar systems using far cheaper hardware. The system cleverly combines a standard 2D imaging radar, which provides dense 360-degree planar data, with a 3-axis gyroscope. By extending the existing SE(2) Direct Radar Odometry (DRO) framework, 3DRO integrates the radar's 2D velocity estimates with the gyroscope's rotational (SO(3)) measurements to compute full six-degree-of-freedom (SE(3)) ego-motion. This fusion allows the system to estimate not just movement in a plane but full 3D translation and rotation, a critical capability for robots and autonomous vehicles navigating complex, uneven terrain.

The breakthrough is in its performance-to-cost ratio. The team rigorously validated 3DRO using the massive Boreas-RT dataset, comprising 643km of real-world driving data. The results demonstrate that this radar-gyroscope combination can achieve odometry accuracy on par with high-end lidar systems. Lidar sensors, while highly precise, are often prohibitively expensive, bulky, and sensitive to adverse weather like fog and rain. In contrast, radar is robust in all weather conditions, more compact, and significantly cheaper. This research, published on arXiv, provides a compelling path toward more affordable and reliable perception stacks for autonomous systems, potentially lowering the barrier to entry for widespread robotics and self-driving car development.