ROAMER: Robust Offroad Autonomy using Multimodal State Estimation with Radar Velocity Integration

TL;DR

ROAMER addresses the challenge of robust, low-latency state estimation for offroad autonomy by integrating forward velocity measurements from a FMCW radar into a LiDAR–inertial graph-based smoother. The approach adds a radar forward-velocity factor to a windowed MAP estimator, deriving velocity residuals from radar Doppler data and fusing them with IMU and LiDAR information; the radar processing includes CFAR filtering and consensus to produce a reliable velocity update. Hardware experiments on a high-speed all-terrain vehicle show that LRI matches LI performance under normal operation and dramatically improves robustness during LiDAR dropout, with notable reductions in RPE and velocity error. The work demonstrates that radar velocity integration can bolster resilience of autonomous navigation in unstructured, high-speed environments, suggesting avenues for richer use of radar data (e.g., radial speed) to further improve multi-axis velocity estimates.

Abstract

Reliable offroad autonomy requires low-latency, high-accuracy state estimates of pose as well as velocity, which remain viable throughout environments with sub-optimal operating conditions for the utilized perception modalities. As state estimation remains a single point of failure system in the majority of aspiring autonomous systems, failing to address the environmental degradation the perception sensors could potentially experience given the operating conditions, can be a mission-critical shortcoming. In this work, a method for integration of radar velocity information in a LiDAR-inertial odometry solution is proposed, enabling consistent estimation performance even with degraded LiDAR-inertial odometry. The proposed method utilizes the direct velocity-measuring capabilities of an Frequency Modulated Continuous Wave (FMCW) radar sensor to enhance the LiDAR-inertial smoother solution onboard the vehicle through integration of the forward velocity measurement into the graph-based smoother. This leads to increased robustness in the overall estimation solution, even in the absence of LiDAR data. This method was validated by hardware experiments conducted onboard an all-terrain vehicle traveling at high speed, ~12 m/s, in demanding offroad environments.

Figures (12)

• Figure 1: RACER robot platform used to conduct offroad autonomy experiments. Figure includes insets from onboard cameras as well as LiDAR map and radar data visualizations at choice locations. Note, the typical environment in the background, consisting of unmaintained dirt paths, arid conditions, and sparse vegetation.
• Figure 2: Visualization of a radar beam, which consists of 12 pixels across 4° of azimuth and 7.5° of elevation. Each pixel consists of a range-doppler image of signal intensities, with size depending on the waveform used. An example of one pixels' range-doppler map can be seen, note this is of a static environment, hence the high intensity returns along 0 doppler.
• Figure 3: Visualization of a selection of radar beam pixels corresponding to particular points in the velocity profile. These pixels are all along -0.5° azimuth and -2.5° elevation so they should correspond very closely with forward velocity. Note the quantity of noise in the data but also how the left-most part of the high-intensity smear corresponds well to forward velocity. Forward velocity here is taken from the LiDAR-inertial odometry onboard the vehicle. Range-doppler images are cropped in size to more easily see the data.
• Figure 4: Redraw of the visualization in richards2013Fundamentals, depicting 1D and 2D cfar kernels used for estimating the interference noise of the cut.
• Figure 5: Visualization of the range-doppler image before and after applying the cacfar filter, where the right column represents all cells in the range-doppler image from the left column which were classified as targets (${\mathcal{H}}_{1}$) rather than interference. Note, that not all images will necessarily have post filter targets in the same doppler column, motivating the usage of consensus to reduce the likelihood of outliers in the radial speed measurement.