Ground-Penetrating Radar
- Also called: GPR, georadar, subsurface radar
- Category: Sensing › Exteroceptive
- Tags: radar, subsurface-sensing, localization, inspection, field-robotics
A sensor that sends radio pulses into the ground and records reflections from buried layers and objects, producing images of the subsurface; in robotics it serves inspection and localization.
Definition
Ground-penetrating radar (GPR) transmits short electromagnetic pulses, at frequencies from tens of megahertz to a few gigahertz, into the ground through an antenna held close to the surface. Wherever the electrical properties of the material change (at a soil layer, a pipe, a void, a rebar or a rock) part of the energy is reflected back to a receiving antenna. The record at one position is a one-dimensional waveform called a trace, or A-scan, showing reflection strength against two-way travel time. Moving the antenna along a line and placing traces side by side gives a B-scan, a two-dimensional image of the subsurface along the path, and a grid of lines can be combined into a 3D volume.
Interpreting GPR data takes care. Travel time converts to depth only once the wave velocity in the material is known, and that velocity depends on the dielectric permittivity, which changes with moisture and soil type. A small buried object appears as a hyperbola in a B-scan, because the antenna detects it before and after passing directly over it; fitting the hyperbola's shape gives both the object's position and the wave velocity. Lower frequencies penetrate deeper but resolve less detail, and conductive ground such as wet clay or salty soil absorbs the signal quickly, limiting depth.
In robotics, GPR appears in two roles. As an inspection payload, it lets robots map rebar in concrete, detect voids under pavement, locate buried utilities or search for landmines. As a localization sensor, it exploits the fact that subsurface structure is distinctive and stable over time, and unaffected by snow, rain, lighting or changes above ground: a vehicle records a GPR map of a route, then later matches live traces against it. Localizing ground-penetrating radar (LGPR), developed at MIT Lincoln Laboratory, demonstrated this for road vehicles, and later work learns encodings of GPR data for matching inside a factor graph. GPR, the radar, should not be confused with Gaussian process regression, a machine learning method that shares the abbreviation.
Why it matters
GPR gives robots a view below the surface that cameras and lidar cannot provide. For inspection it turns robots into tools for infrastructure assessment, and for navigation it offers a localization signal that keeps working when surface appearance changes or weather blinds other sensors.
Key applications
- Vehicle localization against a prior subsurface map
- Robotic inspection of bridge decks, tunnels and concrete for rebar and voids
- Detecting pipes and cables before excavation
- Landmine and unexploded ordnance detection
- Subsurface sounding from planetary rovers
- Archaeological and geological surveys
In practice
A research vehicle drives a route with a GPR array mounted under the chassis, recording traces together with accurate poses to build a prior map. On later drives, the system encodes short windows of live GPR data and compares them with the map to produce position measurements, which enter a factor graph alongside odometry and IMU preintegration factors. Engineers keep the antenna height consistent, remove the direct transmitter-to-receiver signal and the ground-surface reflection from each trace before matching, and check how much the signal shifts after rain changes the soil moisture.
Common misconceptions
- GPR shows a picture of what is buried — it records reflections over time; depth and shape must be inferred, and hyperbolas, multiple reflections and noise need interpretation
- Depth can be read straight from the data — converting travel time to depth requires the wave velocity, which varies with soil type and moisture
- GPR works the same in any ground — conductive soils such as wet clay absorb the signal and can reduce penetration to very little
- GPR in a robotics paper always means the radar — the same abbreviation is used for Gaussian process regression, an unrelated learning method
Who should know this
- Field and infrastructure inspection robotics engineers
- Autonomous vehicle localization researchers
- Geophysics and survey engineers working with robots
- Planetary rover scientists and engineers
Related technologies
- Localizing Ground-Penetrating Radar (LGPR)
- Ultra-Wideband Radar
- Electromagnetic Induction Sensors
- Migration Imaging
- Factor Graph Localization
Related terms
Related projects
- Cartographer: Google's real-time 2D and 3D SLAM system for multi-platform robotics
- OpenDroneMap: Open-source photogrammetry toolkit that turns drone images into maps and 3D models
- Nav2: ROS 2 navigation stack with path planning and obstacle avoidance for mobile robots
- AprilTag: The standard visual fiducial marker system for robot localization and pose estimation
- GLIM: GPU-accelerated 3D lidar SLAM framework for any range sensor, from AIST
- M2DGR: Multi-modal, multi-scenario SLAM dataset recorded by a ground robot (RA-L 2022)
Videos
- How ground-penetrating radar works: CBC News