Learning Surface Terrain Classifications from Ground Penetrating Radar

TL;DR

The paper investigates terrain classification for mobile robots in challenging environments using Ground Penetrating Radar (GPR) instead of relying solely on RGB vision. It proposes a processing pipeline that extracts direct-wave radargram segments, applies multiple classifiers (2D CNNs like AlexNet/ResNet101 and a 1D CNN, plus a variational autoencoder for deep clustering) to label surface terrains, and examines time-series length. A new dataset with four terrain types is collected on a mobile robot platform, and results show that GPR signals carry discriminative surface-terrain information, with direct-wave portions performing best for surface classification. The work also demonstrates a multimodal mapping example that fuses GPR-derived terrain labels with RGB-based maps, highlighting the practical potential of GPR to augment surface terrain understanding in robotics.

Abstract

Terrain classification is an important problem for mobile robots operating in extreme environments as it can aid downstream tasks such as autonomous navigation and planning. While RGB cameras are widely used for terrain identification, vision-based methods can suffer due to poor lighting conditions and occlusions. In this paper, we propose the novel use of Ground Penetrating Radar (GPR) for terrain characterization for mobile robot platforms. Our approach leverages machine learning for surface terrain classification from GPR data. We collect a new dataset consisting of four different terrain types, and present qualitative and quantitative results. Our results demonstrate that classification networks can learn terrain categories from GPR signals. Additionally, we integrate our GPR-based classification approach into a multimodal semantic mapping framework to demonstrate a practical use case of GPR for surface terrain classification on mobile robots. Overall, this work extends the usability of GPR sensors deployed on robots to enable terrain classification in addition to GPR's existing scientific use cases.

Figures (8)

• Figure 1: Clearpath Husky robot with a GPR mounted on the back. The radargram, which represents the permittivity of the surface and subsurface materials sensed by the GPR as a time series, is shown in image form in the bottom left.
• Figure 2: 60 x 32 pixel slice of a GPR radargram direct wave for four terrain types: a. asphalt, b. grass, c. sand, and d. sidewalk.
• Figure 3: The radar propagation from GPR pulses has a unique property where it travels through the air and the surface of the ground (direct wave, blue), and also through the subsurface material itself (reflected wave, red). The radargram on the right shows how these different received waves appear in a GPR radargram.
• Figure 4: Our custom Clearpath Husky robotic sensor platform equipped with a GPR, a GPS, an RGBD camera, and wheel encoders.
• Figure 5: The five time series lengths used for the experiment shown on a single radargram of sandy terrain. Note that the pattern distinct to the terrain type makes itself more apparent as the number of traces increases.