UGVs in Precision Agriculture: Mapping, Sampling, and Monitoring in the Field

TL;DR: UGV precision agriculture combines unmanned ground vehicles, RTK-GNSS, and multispectral or LiDAR payloads to map fields, collect soil and plant samples, and monitor crop health at row-level resolution. Compared to UAVs, UGVs operate under the canopy, carry heavier sensors, and run longer missions. Platforms like the Leo Rover let research teams prototype mapping, sampling, and monitoring workflows in ROS 2 without building a chassis from scratch.

Why UGVs matter for precision agriculture

Precision agriculture depends on dense, georeferenced data: soil moisture, NDVI, plant height, disease pressure, weed location. Satellites and drones cover area fast, but they look down from above. They miss what happens beneath the canopy, at the stem, or in the top few centimeters of soil. That gap is where UGV precision agriculture earns its place.

An unmanned ground vehicle drives along rows, holds position with RTK-GNSS, and carries payloads too heavy or too power-hungry for a small UAV. It can stop, lower a probe, take a measurement, and continue. It can repeat the same trajectory week after week, giving you time-series data with consistent geometry. For researchers and integrators building phenotyping rigs, weed detection systems, or autonomous scouting platforms, the ground vehicle is often the only sensor carrier that fits the task.

Where UGVs outperform aerial platforms

The decision between UAV and UGV is not exclusive, but each has clear strengths. Before choosing, it helps to look at the practical differences:

• Endurance: a small electric UGV can run for hours on one battery charge, while a comparable quadcopter typically lasts 20-40 minutes.
• Payload: ground robots carry multispectral cameras, hyperspectral sensors, LiDAR, soil probes, and manipulators together, without the strict flight-weight limits of UAVs.
• Proximity: sensors operate centimeters from leaves, soil, or fruit, which improves spatial resolution and reduces atmospheric noise.
• Repeatability: with RTK and a fixed path plan, nearly the same wheel tracks are followed each visit, helping isolate biological change from sensor placement noise.
• Regulation: ground operation avoids most aviation rules, which simplifies field trials on private farmland.

How UGVs map agricultural fields

Mapping is the foundation layer. Without an accurate map, sampling locations drift and monitoring time-series lose their value. In UGV precision agriculture, mapping usually combines three data sources: RTK-GNSS for absolute position, wheel odometry and IMU for short-term motion, and LiDAR or stereo vision for local geometry.

A typical ROS 2 stack uses robot_localization to fuse GNSS, wheel odometry, and IMU into an EKF, slam_toolbox or rtabmap_ros for 2D/3D SLAM, and nav2 for path planning along crop rows. RTK corrections over NTRIP bring horizontal accuracy to roughly 1-2 cm in open sky, which is enough to align measurements between visits across a growing season.

Row-following and headland turns

Open fields look uniform to a global planner, so many agricultural UGVs use a row-detection layer on top of the map. LiDAR or depth cameras detect the crop walls on either side, and a local controller keeps the robot centered. At the end of a row, the planner switches to a headland turn defined in field coordinates. The Leo Rover, for example, can be used in research deployments where ROS 2 nodes handle row tracking and a high-level mission node sequences rows from a GeoJSON field boundary.

How UGVs handle sampling and physical interaction

Sampling is what separates a UGV from a flying scout. A robot that can stop and touch the ground or the plant collects data no camera can produce. Common sampling payloads include soil moisture probes, electrical conductivity sensors, leaf clip spectrometers, and small coring tools for laboratory analysis.

The mechanical integration usually involves a small electric actuator or a lightweight arm mounted on the rover deck. ROS 2 action servers fit this pattern well: a TakeSample action receives a target waypoint, the navigation stack drives there, the manipulation node runs the probe cycle, and the result (a measurement plus a timestamp and RTK fix) is logged into a rosbag or pushed to a database.

Real research examples include the Thorvald platform from Saga Robotics for strawberry UV treatment, the BoniRob from Bosch and Amazone for weed and phenotyping work, and the TerraSentia from EarthSense for under-canopy phenotyping of corn and soybean. These projects show that field robotics scales from research prototypes to commercial deployment when the sampling and mapping pipelines are solid.

How UGVs perform crop monitoring with multispectral and RTK sensors

Monitoring is the repeat-visit task: drive the same path, collect the same data, compare against last week. The sensor stack typically includes a multispectral camera (for example MicaSense RedEdge or Sentera 6X), an RGB camera for ground-truth and weed ID, sometimes a thermal camera, and an RTK-GNSS receiver for georeferencing every frame.

Multispectral imaging gives you NDVI, NDRE, and other vegetation indices at sub-centimeter ground sample distance when the camera sits roughly 1 m above the canopy. That is about one to two orders of magnitude finer than what a UAV at 30 m altitude provides, which lets you separate individual plants and, in suitable conditions, individual leaves.

A practical ROS 2 monitoring pipeline

To turn raw frames into a useful crop map, the pipeline usually follows these steps:

1. Trigger the multispectral camera at fixed distance intervals using odometry, not time, to keep overlap consistent.
2. Tag each frame with the RTK fix from /gps/fix and the robot pose from the EKF.
3. Apply radiometric calibration using a reference panel image captured at mission start.
4. Stitch frames into an orthomosaic with a tool such as OpenDroneMap or Pix4D, using the RTK positions as high-accuracy geotags and adding ground control or check points when accuracy must be validated.
5. Compute vegetation indices and export as GeoTIFF for analysis in QGIS or a farm management system.

For teams building this stack, the Leo Rover provides a ROS 2-capable base with documented payload mounts, payload capacity of up to 5 kg, and integration examples for GNSS, LiDAR, and cameras. More details on field-ready configurations are available on the Leo Rover page and in the broader research applications overview.

What hardware constraints define a field-ready UGV

Field deployment is rough on robots. Dust, sun, mud, dew, and uneven ground all matter. When evaluating or designing an agricultural UGV, a few specifications carry most of the weight:

• Ingress protection: at least IP54 on the main electronics enclosure for outdoor use in dew and light rain.
• Ground clearance: enough to clear furrows and small residue, typically 10 cm or more for row crops at early growth stages.
• Wheel traction: pneumatic or knobbed tires for soft soil; tracked variants for very wet conditions.
• Power budget: sensors, compute, and motors together; a multispectral camera plus a small NVIDIA Jetson plus 4G modem can draw 30-50 W continuous.
• Open software stack: ROS 2 support shortens integration time for new sensors and custom mission logic.

The Leo Rover platform was designed around these constraints for research and light-duty field robotics, and similar logic applies to any UGV you evaluate for precision agriculture work.

FAQ

What is a UGV in precision agriculture?

A UGV (unmanned ground vehicle) in precision agriculture is a wheeled or tracked robot that drives through fields autonomously or semi-autonomously to collect data, take samples, or apply treatments. It complements UAVs by working at canopy level or below.

How accurate is RTK-GNSS on a small agricultural robot?

With a clear sky view and NTRIP corrections from a nearby base station, RTK-GNSS typically reaches 1-2 cm horizontal accuracy. Accuracy degrades under tree cover or near tall structures due to multipath and reduced satellite visibility.

Can a UGV replace a drone for crop monitoring?

Not in most cases. Drones cover large areas quickly; UGVs cover smaller areas at much higher spatial resolution and can sample physically. Many research programs use both, with UAVs for scouting and UGVs for ground-truth and intervention.

Which sensors are most common on agricultural UGVs?

Multispectral cameras, RGB cameras, LiDAR, thermal cameras, RTK-GNSS receivers, IMUs, and soil probes are the typical payload. The exact combination depends on the crop and the research question.

Is ROS 2 a good fit for field robotics?

Yes. ROS 2 provides DDS-based middleware that can be configured for real-time-oriented systems, mature navigation and SLAM stacks, and a large ecosystem of sensor drivers. Many modern research UGVs, including the Leo Rover, offer ROS 2 support.

How long can a small UGV operate in the field?

Runtime depends on terrain, speed, and payload load, but small electric platforms commonly run 2-6 hours on one battery, with swappable batteries extending missions further.

What are realistic research use cases to start with?

Row-following with RTK waypoints, NDVI mapping with a multispectral camera, soil moisture sampling at predefined grid points, and weed detection with an RGB camera and a trained vision model are all reachable starting projects.

To see how Fictionlab’s platforms fit into mapping, sampling, and monitoring workflows, visit the agriculture applications page.