Agricultural robot

Agricultural robot – definition

An agricultural robot is a robotic system designed to perform sensing, navigation, inspection, measurement, or physical work in agricultural environments. In mobile robotics, the term usually refers to an unmanned ground vehicle (UGV) that operates in fields, orchards, greenhouses, farms, or other semi-structured outdoor areas.

An agricultural UGV combines a mobile base, onboard computing, sensors, software, and task-specific payloads. It may be teleoperated, supervised by an operator, or autonomous within defined operational limits. Autonomy is not an inherent property of the platform. It requires localisation, perception, path planning, motion control, safety logic, and a validated operating procedure.

For ROS-based systems, an agricultural robot is commonly modelled as a robot with a coordinate-frame tree, sensor topics, velocity commands, odometry, and a navigation stack. ROS conventions for coordinate frames are defined by REP 103, while mobile-platform frame relationships such as map, odom, and base_link are described in REP 105.

Typical agricultural UGV tasks

Agricultural robots are used where repeated data collection or controlled movement is required over large or difficult-to-access areas. The task determines the required sensors, payload capacity, environmental protection, and navigation method.

Crop and field inspection – collecting RGB, thermal, multispectral, depth, or LiDAR data for crop-condition analysis.
Row following – navigating between crop rows using camera, LiDAR, GNSS/RTK, wheel odometry, or sensor fusion.
Phenotyping and research – recording plant geometry, canopy properties, growth stages, and environmental measurements.
Soil and environmental monitoring – carrying sensors for moisture, temperature, air quality, or soil sampling.
Targeted field operations – positioning a task-specific payload, such as an inspection camera or sampling tool, at defined locations.

These applications differ from road-vehicle autonomy. An agricultural UGV normally operates at low speed, in a restricted area, and under an operational safety procedure. However, uneven terrain, vegetation, dust, changing illumination, wheel slip, and GNSS degradation make localisation and perception technically demanding.

ROS 2 architecture for an agricultural robot

ROS 2 provides communication, message definitions, launch tools, parameter handling, and integration patterns for an agricultural UGV. It does not provide complete agricultural autonomy by itself. A robot application must integrate hardware drivers and higher-level software components.

A common ROS 2 data flow contains the following interfaces:

sensor_msgs/msg/Image and sensor_msgs/msg/CameraInfo for RGB or depth cameras.
sensor_msgs/msg/LaserScan or sensor_msgs/msg/PointCloud2 for 2D and 3D LiDAR data.
sensor_msgs/msg/Imu for angular velocity, linear acceleration, and orientation data.
sensor_msgs/msg/NavSatFix for GNSS position data.
nav_msgs/msg/Odometry for estimated robot pose and velocity in the odom frame.
geometry_msgs/msg/Twist on /cmd_vel for linear and angular velocity commands.

Message definitions are maintained in the official ROS 2 interface packages, including sensor_msgs, nav_msgs, and geometry_msgs. The exact topic names are implementation-specific, but frame IDs and timestamps must be consistent across all sensor drivers.

Localisation and navigation in agricultural environments

An agricultural robot needs an estimate of its position relative to a local map, field reference system, crop row, or planned route. The choice of localisation method depends on the environment and required accuracy.

Wheel odometry estimates motion from wheel rotation. It is useful for short-term motion estimation but accumulates error when wheels slip on soil, gravel, grass, or mud.
IMU fusion improves orientation and short-term motion estimates. It does not independently provide globally bounded position.
GNSS/RTK provides global positioning where satellite visibility and correction data are available. It may be degraded near trees, buildings, greenhouses, or terrain obstructions.
LiDAR or visual SLAM estimates motion from observed geometry or visual features. Performance depends on scene structure, lighting, vegetation movement, and sensor placement.
Row-following perception estimates a traversable corridor from plant rows, soil boundaries, or artificial markers.

For autonomous navigation, the ROS 2 Navigation Stack, commonly called Nav2, can generate paths and velocity commands from a map, costmaps, localisation data, and robot footprint configuration. Nav2 documentation defines its architecture, lifecycle-managed nodes, planners, controllers, and behaviour-tree execution model in the official Nav2 documentation.

Leo Rover and Raph Rover in agricultural robotics

Leo Rover is suitable for prototyping agricultural sensing, row-following, teleoperation, and ROS 2 navigation experiments. Its standard compute unit is based on a Raspberry Pi, and ROS 2 compatibility depends on the installed software image and supported packages. Leo Rover uses a four-wheel skid-steer configuration and has an IP55 rating. These platform characteristics are documented by Fictionlab in the Leo Rover technical documentation.

For field work, Leo Rover requires additional integration. A practical system may include a LiDAR or depth camera for obstacle detection, an IMU for attitude estimation, a GNSS/RTK receiver for field coordinates, and a payload-specific sensor. The robot is not an autonomous agricultural vehicle out of the box. Navigation, sensor calibration, obstacle behaviour, emergency-stop procedures, and field validation must be implemented for the intended task.

Raph Rover is more appropriate when the application requires a larger payload, additional battery capacity, a larger sensor mast, or heavier research equipment. It remains a UGV platform rather than a complete agricultural robot. The final system architecture depends on payload mass, terrain, centre of gravity, compute requirements, and the safety case.

Example ROS 2 configuration

The following simplified configuration shows interfaces commonly used when a field robot fuses wheel odometry and IMU data. Topic names must match the installed drivers and the robot frame tree.

ekf_filter_node:
  ros__parameters:
    frequency: 30.0
    two_d_mode: true
    odom0: /wheel/odometry
    odom0_config: [true, true, false,
                   false, false, true,
                   true, false, false,
                   false, false, true,
                   false, false, false]
    imu0: /imu/data
    imu0_config: [false, false, false,
                  false, false, true,
                  false, false, false,
                  false, false, true,
                  false, false, false]
    world_frame: odom

The frequency value above is an application parameter, not a universal requirement. It must be selected according to sensor update rates, processor load, vehicle speed, and controller requirements.

Safety and standards context

Agricultural autonomy must be assessed as a system-level safety problem. The assessment includes unintended motion, loss of communications, sensor faults, actuator faults, emergency-stop behaviour, obstacle handling, and operation near people.

Relevant standards include ISO 18497-1:2024, which replaced ISO 18497:2018 and addresses safety of partially automated, semi-autonomous and autonomous agricultural machinery, and ISO 25119, which specifies safety-related parts of control systems for agricultural and forestry machinery. These standards do not replace application-specific risk assessment, local regulations, or controlled field testing.

Agricultural robot