Ground Penetrating Radar-Assisted Multimodal Robot Odometry Using Subsurface Feature Matrix

TL;DR

This work tackles GPS-denied robot localization by leveraging subsurface information from ground-penetrating radar (GPR). It introduces a subsurface feature matrix (SFM) derived from GPR B-scan data and a peak-based matching strategy to estimate GPR-driven travel, which is fused with IMU and wheel encoder data in a factor-graph. The proposed approach achieves a GPR-component RMSE of $2.23$ m and an overall odometry RMSE of $0.568$ m on the CMU-GPR dataset, outperforming several baselines and operating in real time. The method eliminates the need for training, enhances robustness to environmental changes, and shows strong potential for subsurface-assisted robotic navigation.

Abstract

Localization of robots using subsurface features observed by ground-penetrating radar (GPR) enhances and adds robustness to common sensor modalities, as subsurface features are less affected by weather, seasons, and surface changes. We introduce an innovative multimodal odometry approach using inputs from GPR, an inertial measurement unit (IMU), and a wheel encoder. To efficiently address GPR signal noise, we introduce an advanced feature representation called the subsurface feature matrix (SFM). The SFM leverages frequency domain data and identifies peaks within radar scans. Additionally, we propose a novel feature matching method that estimates GPR displacement by aligning SFMs. The integrations from these three input sources are consolidated using a factor graph approach to achieve multimodal robot odometry. Our method has been developed and evaluated with the CMU-GPR public dataset, demonstrating improvements in accuracy and robustness with real-time performance in robotic odometry tasks.

Figures (7)

• Figure 1: Illustration of our SFM based multimodal odometry algorithm.
• Figure 2: A simplified visualization of GPR working principle from Gjjpaper1.
• Figure 3: The algorithm flow chart of the paper
• Figure 4: Sample $B_t(x,y)$ in (a), $B'_t(x,y)$ in (b), and $S_{t}$ in (c)
• Figure 5: Efficacy of the GPR odometry with SFM: (a) trajectory estimated from individual modalities including total station (ground truth), wheel encoder and GPR. (b) RMSE comparison between the GPR and the wheel encoder modalities.