Are Doppler Velocity Measurements Useful for Spinning Radar Odometry?

TL;DR

This work demonstrates that Doppler velocity measurements can meaningfully enhance spinning radar odometry. By exploiting a triangular modulation scheme, the authors extract Doppler velocities from consecutive azimuths without data association and integrate them into both ICP-based and direct-velocity odometry pipelines. Across four challenging driving environments totaling over 110 km, Doppler-enabled methods maintain functional odometry in geometrically degenerate settings (e.g., tunnels) where traditional baselines fail, with the best performance when fusing Doppler, gyroscope data, and ICP. The results highlight the practical potential of Doppler radar to improve robust navigation in autonomous vehicles without additional hardware costs, and suggest avenues for further integration with mapping and dynamic-object handling.

Abstract

Spinning, frequency-modulated continuous-wave (FMCW) radars with 360 degree coverage have been gaining popularity for autonomous-vehicle navigation. However, unlike `fixed' automotive radar, commercially available spinning radar systems typically do not produce radial velocities due to the lack of repeated measurements in the same direction and the fundamental hardware setup. To make these radial velocities observable, we modified the firmware of a commercial spinning radar to use triangular frequency modulation. In this paper, we develop a novel way to use this modulation to extract radial Doppler velocity measurements from consecutive azimuths of a radar intensity scan, without any data association. We show that these noisy, error-prone measurements contain enough information to provide good ego-velocity estimates, and incorporate these estimates into different modern odometry pipelines. We extensively evaluate the pipelines on over 110 km of driving data in progressively more geometrically challenging autonomous-driving environments. We show that Doppler velocity measurements improve odometry in well-defined geometric conditions and enable it to continue functioning even in severely geometrically degenerate environments, such as long tunnels.

Figures (7)

• Figure 1: Top: The estimated vehicle velocity using a baseline and our Doppler-enabled method. Bottom: Our data collection platform Boreas in front of the Burlington Bay James N. Allan Skyway. The skyway has few reliable geometric features, resulting in the baseline estimation failure.
• Figure 2: A graphical representation of the shift in the frequency between the transmitted signal in blue and the received signal in orange for an FMCW radar with a triangular modulation pattern of slope $S$. The top shows a frequency shift $\Delta f_t$ resulting from a time delay $\Delta t$ induced by the signal traveling some distance. The bottom shows a frequency shift $\Delta f_d$ resulting from the Doppler effect of an object moving away from the radar.
• Figure 3: Static features measured using a sawtooth (top) and triangular (bottom) modulated moving radar. The alternating Doppler-induced shift produced by the triangular signal modulation pattern can be seen as a 'zig-zag' in the intensity returns of continuous, flat features in the bottom image.
• Figure 4: The per-scan Doppler velocity estimation pipeline. First, consecutive pairs of raw radar intensity signals $\mathrm{z}_k, \, k \in \{0, N\}$ from azimuths $\phi_k$ are loaded into the radial velocity extractor. Each pair is filtered and a cross-correlation is run to estimate a radial velocity $u_j, \, j \in \{1/2, N-1/2\}$. Radial velocities for the entire scan $\mbf{u}$ are passed through RANSAC and the inliers $\mbf{u}_R$ used in a robust least-squares algorithm to estimate the ego-velocity $\mbf{v}$.
• Figure 5: Groundtruth and estimated radial velocities projected onto a radar pointcloud in a static and dynamic scene. RANSAC-rejected velocities are coloured in orange. The vehicle orientation is shown as a green arrow.