Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Dhruv2012/Autonomous-Farm-Robot/llms.txt

Use this file to discover all available pages before exploring further.

AGRIBOT includes a complete Gazebo simulation environment featuring a 50 × 50 m farm field world with textured grass and pavement tiles, a grid of cylindrical CropWithWeed plant models, an entry gate pole, and a surrounding scene of pine trees, oak trees, and lamp posts. The robot model itself carries simulated GPS (NEO-M8N style via libhector_gazebo_ros_gps.so), IMU, magnetometer, and a 640 × 480 RGB camera — all with configurable Gaussian noise — making the simulation closely representative of real-world sensor conditions.

​
Available Launch Files

The agribot_description and agribot_control packages expose five launch files that cover every stage of the simulation workflow, from bare URDF loading through to full autonomous drive.
Loads agribot_body.urdf.xacro onto the ROS parameter server and starts robot_state_publisher, joint_state_publisher, and RViz with a default display configuration. Use this for URDF inspection and RViz-only visualization without a running Gazebo instance.Nodes started: joint_state_publisher, robot_state_publisher, rviz
Spawns AGRIBOT (agribot_sensors.urdf.xacro) into an empty Gazebo world using gazebo_ros/empty_world.launch. Also publishes a static TF from base_link to base_footprint and fires a fake joint calibration signal on /calibrated.Nodes started: gazebo, spawn_model, tf_footprint_base, fake_joint_calibration
Spawns AGRIBOT into the custom Farm.world Gazebo scene. Also starts joint_state_publisher, robot_state_publisher, a static TF from odom to map, and includes agribot_control.launch automatically. The GPS converter and data manipulation nodes are also launched in separate xterm terminals from this file.Nodes started: gazebo, joint_state_publisher, robot_state_publisher, controller_spawner, gps_converter, sensor_data, plus a static TF broadcaster
Starts joint_state_publisher and robot_state_publisher only, without launching RViz or Gazebo. Use this as a lightweight base when you want to bring up the robot state topics and then open your own RViz session manually.Nodes started: joint_state_publisher, robot_state_publisher
Loads the PID controller configuration from agribot_control/config/agribot_control.yaml and spawns five controllers under the /agribot namespace: joint_state_controller, rightwheelR_effort_controller, leftwheelR_effort_controller, rightwheelF_effort_controller, and leftwheelF_effort_controller. Also loads agribot_sensors.urdf.xacro onto the parameter server.Nodes started: controller_spawner
Use the commands below to launch the most common simulation configurations: 
# Spawn in empty world
roslaunch agribot_description agribot_gazebo_simulation.launch

# Spawn in farm world (recommended — includes controllers and GPS nodes)
roslaunch agribot_description my_world.launch

# Visualize in RViz
roslaunch agribot_description agribot_rviz_display.launch

​
Farm World Description

The Farm.world SDF file defines the complete simulation environment. Understanding its structure helps when extending or debugging the scene. Terrain layout. The ground plane is a 50 × 50 m static model (farm_field_0_0) composed of 100 tiled 5 × 5 m visual panels. Most panels use the vrc/grass material, while an interior band of panels uses the vrc/pavement material (Gazebo/Grey) — creating a visual driving corridor distinct from the surrounding field grass. CropWithWeed models. The farm field is populated with a grid of CropWithWeed SDF objects, each a cylinder of radius 0.055 m and length 0.2 m placed 0.115 m above ground. Their visual material (CropWithWeed/Diffuse) is driven by a custom OGRE script (CropWithWeed.material) that applies a plant texture PNG. The cylinder geometry provides both a collision shape for the physics engine and a surface for the camera-based CNN classifier to segment. Simulated sensors on the robot. The AGRIBOT model embedded in Farm.world runs the following Gazebo plugins:
PluginTopicUpdate Rate
libhector_gazebo_ros_gps.so/agribot/fix, /agribot/fix_velocity5 Hz
libhector_gazebo_ros_imu.so/agribot/imu200 Hz
libhector_gazebo_ros_magnetic.so/magnetic5 Hz
libgazebo_ros_camera.so/agribot/camera/image_raw30 Hz
libgazebo_ros_skid_steer_drive.so/agribot/cmd_vel, /agribot/odom100 Hz
The GPS reference origin is set to latitude 21.161289°, longitude 72.787022° (SVNIT campus, Surat, India). Gaussian noise of σ = 0.1 is applied to all three GPS axes and drift of 5.0 m is configured to replicate real-world GPS inaccuracy. Map configuration. The maps/farm.yaml file defines the occupancy grid used with the optional map_server node: 
image: field.png
resolution: 0.1
origin: [-10.0, -10.0, 0.0]
negate: 0
occupied_thresh: 0.65
free_thresh: 0.196
The map origin is set 10 m west and 10 m south of the Gazebo world origin, aligning the occupancy grid with the farm field tile layout.

​
Teleoperation for Manual Testing

Before running the autonomous drive nodes, you can drive AGRIBOT manually to verify that the controllers, motor effort publishers, and skid-steer drive plugin are working correctly. Open a new terminal and run one of the following: 
# Keyboard teleoperation
rosrun key_teleop key_teleop.py

# Joystick teleoperation
rosrun joy_teleop joy_teleop.py
Both methods publish geometry_msgs/Twist messages to /agribot/cmd_vel, which the libgazebo_ros_skid_steer_drive.so plugin consumes to apply torque to all four wheel joints. Confirm the robot moves in the expected direction before starting the GPS navigation stack.

​
Recording and Replaying Sensor Data

Pre-recorded ROS bag files are included in Documents/bag_files/. These bags capture all sensor topics (GPS fix, IMU, magnetometer, camera, odometry) from actual simulation runs and can be replayed to test the navigation and classification nodes without running a live Gazebo instance. 
# Replay a recorded session
rosbag play Documents/bag_files/data.bag

# Record your own session
rosbag record -a -O my_session.bag
Replaying a bag at full speed is useful for tuning the Kalman filter parameters or the CNN inference pipeline offline. Use rosbag play --clock and set use_sim_time:=true on the ROS parameter server if any nodes depend on the ROS clock.
Use the Mapviz tile_map plugin with a satellite tile source (such as the OpenStreetMap or Bing Maps WMS endpoint) to overlay the GPS trace on a real aerial map during simulation. This makes it straightforward to correlate the simulated farm field geometry with a real field layout when preparing for hardware deployment.

Build docs developers (and LLMs) love

Get started for freeTalk to us