MarsLGPR: Mars Rover Localization with Ground Penetrating Radar

TL;DR

This work tackles the challenge of GPS-denied rover localization on Mars by leveraging Ground Penetrating Radar (GPR) as an auxiliary sensing modality. It introduces GPRFormer, a transformer-based network that estimates relative 1D displacement from GPR B-scans and integrates these predictions into an Extended Kalman Filter alongside wheel odometry and IMU data. The MarsLGPR dataset provides the first Mars-analog GPR localization data, and experiments on CMU-GPR and MarsLGPR show that GPRFormer improves localization accuracy, especially under high wheel slip, while maintaining real-time performance. Overall, GPR-based subsurface features emerge as a valuable complement to traditional odometry for robust planetary navigation.

Abstract

In this work, we propose the use of Ground Penetrating Radar (GPR) for rover localization on Mars. Precise pose estimation is an important task for mobile robots exploring planetary surfaces, as they operate in GPS-denied environments. Although visual odometry provides accurate localization, it is computationally expensive and can fail in dim or high-contrast lighting. Wheel encoders can also provide odometry estimation, but are prone to slipping on the sandy terrain encountered on Mars. Although traditionally a scientific surveying sensor, GPR has been used on Earth for terrain classification and localization through subsurface feature matching. The Perseverance rover and the upcoming ExoMars rover have GPR sensors already equipped to aid in the search of water and mineral resources. We propose to leverage GPR to aid in Mars rover localization. Specifically, we develop a novel GPR-based deep learning model that predicts 1D relative pose translation. We fuse our GPR pose prediction method with inertial and wheel encoder data in a filtering framework to output rover localization. We perform experiments in a Mars analog environment and demonstrate that our GPR-based displacement predictions both outperform wheel encoders and improve multi-modal filtering estimates in high-slip environments. Lastly, we present the first dataset aimed at GPR-based localization in Mars analog environments, which will be made publicly available at https://umfieldrobotics.github.io/marslgpr.

Figures (13)

• Figure 1: Our mobile robot collection platform in a Mars analog environment. The proposed method demonstrates a novel approach to relative displacement estimation with Ground Penetrating Radar, which then informs a filtering-based state estimation algorithm.
• Figure 2: Consecutive GPR samples have overlapping beam spreads, which cause them to detect the same object across multiple returns. Because of this, the resulting GPR B-scan image may not intuitively represent the geometry of objects underground. However, we can use this property of the sensor to predict relative pose translation using deep learning.
• Figure 3: Our GPRFormer network takes in a GPR B-scan with $k$ consecutive traces, performs a linear encoding, adds learned positional embeddings, and then passes the encoded tokens through a compact transformer hassani2021. The output from the network is an estimate of relative displacement ($\Delta d$) that the vehicle traveled while collecting the GPR traces in the input.
• Figure 4: A comparison of a raw B-scan containing $k = 10$ traces (a.) and the filtered result after applying background removal, dewow filtering, SEC gain, and wavelet denoising (b.). Note that the direct wave bands are removed and the deeper features (more rightwards) are more pronounced.
• Figure 5: An overview of the dataset organization. There are 50 sequences, totaling around 1TB of data.