MultiRobot Exploration

TFG John Alejandro Gutiérrez: Exploración Multi-Robot en Easynav.

Dependencies & Prerequisites

This package has been developed and strictly tested on Ubuntu 24.04 with ROS 2 Kilted. To successfully build and run the multi-robot exploration nodes, you need the following system and ROS 2 dependencies:

• ROS 2 Base: rclcpp, nav_msgs, geometry_msgs, visualization_msgs
• TF2 Ecosystem: tf2_ros, tf2_geometry_msgs (for global and local frame transformations).
• BehaviorTree.CPP (v4): Used for the decision-making and execution flow of the exploration nodes.
• OpenCV (cv2): Core requirement for the image-processing morphological algorithms used in frontier detection and maps multiplexor.

Installation & Build Instructions

1. Create a workspace and clone the repository:

mkdir -p ~/easynav_mre_ws/src
cd ~/easynav_mre_ws/src
git clone https://github.com/jagutic/easynav_multirobot_exploration easynav_multirobot_exploration

2. Install dependencies automatically using rosdep: Make sure you have sourced your ROS 2 installation first.

cd ~/multi_ws
rosdep update
rosdep install --from-paths src --ignore-src -r -y

(Manual fallback if rosdep fails: sudo apt install libopencv-dev ros-humble-behaviortree-cpp)

3. Build the packages:

colcon build --symlink-install

4. Source the workspace:

source install/setup.bash

EasyNav MultiRobot Exploration Tutorial

Start the simulator

ros2 launch easynav_multirobot_exploration multirobot_sim.launch.py

Use gui:=false to start gazebo headless. Use world:=hospital or world:=maze to use different simulations. Use size:=small or size:=big to use simulations on different size.

Launch SLAM robot mapping and localization

Use use_sim_time:=true or use_sim_time:=false to alternate between simulation and real robots. By default is set to 'true' (simulation).

# Terminal 1 (robot r1)
ros2 launch easynav_multirobot_exploration slam_namespaced.launch.py namespace:=r1

# Terminal 2 (robot r2)
ros2 launch easynav_multirobot_exploration slam_namespaced.launch.py namespace:=r2

Start EasyNav for each robot to be able to navigate

Use use_sim_time:=true or use_sim_time:=false to alternate between simulation and real robots. By default is set to 'true' (simulation).

# Terminal 1 (robot r1)
ros2 run easynav_system system_main \
--ros-args \
--params-file ./src/easynav_multirobot_exploration/easynav_multirobot_exploration/config/navigation_costmap.params.yaml \
-r __ns:=/r1 \
-r /tf:=tf -r /tf_static:=tf_static

# Terminal 2 (robot r2)
ros2 run easynav_system system_main \
--ros-args \
--params-file ./src/easynav_multirobot_exploration/easynav_multirobot_exploration/config/navigation_costmap.params.yaml \
-r __ns:=/r2 \
-r /tf:=tf -r /tf_static:=tf_static

Launch Exploration behaviour for each robot

Use use_sim_time:=true or use_sim_time:=false to alternate between simulation and real robots. By default is set to 'true' (simulation).

# Terminal 1 (robot r1)
ros2 launch easynav_multirobot_exploration exploration.launch.py namespace:=r1

# Terminal 2 (robot r2)
ros2 launch easynav_multirobot_exploration exploration.launch.py namespace:=r2

Launch RViz instances for visualization

Use use_sim_time:=true or use_sim_time:=false to alternate between simulation and real robots. By default is set to 'true' (simulation).

# Terminal 1 (robot r1)
ros2 launch easynav_multirobot_exploration rviz_namespaced.launch.py namespace:=r1

# Terminal 2 (robot r2)
ros2 launch easynav_multirobot_exploration rviz_namespaced.launch.py namespace:=r2

Tips & Gotchas

• Namespaces everywhere: ensure all relative topics in your parameters (e.g., scan_raw) are resolved under each robot namespace (e.g., /r1/scan_raw).
• Do not over-remap: only remap /tf and /tf_static to relative topic names if each robot maintains its own TF tree.
• Frames: with tf_prefix set, refer to frames as r1/base_link, r1/odom, etc.
• Discovery: if you have separate networks or want isolation, consider different ROS_DOMAIN_ID per fleet. Otherwise, keep the same domain to allow shared visualization.

System Architecture

EasyNav plugins

• Lazy Localizer: A lightweight localization plugin designed to reuse SLAM published TFs. Ensures that different localization sources do not collide or create jumps in the transform tree during the exploration process, where both SLAM and Navigation nned to be executed at the same time.

• Multiplexor MapsManager: Works under Costmap MapsManager. Receive all maps to mux through the topic passed as parameter. Uses OpenCV to handle the merging of multiple OccupancyGrid into a global map. This allows robots to share their knowledge of the world. Set this new muxed map as static map for all easynav system.

• Frontier MapsManager: Works over Costmap MapsManager. Using dynamic map saved at navstate (blackboard), extract frontier between free space and the rest of the map using OpenCV morphological filters. Erase noise for better use. Publishes frontier points for behaviours to use it, and saves it in navstate (blackboard) for debugging.

Exploration Behaviour

The autonomous decision-making flow is orchestrated using a reactive Behavior Tree (powered by BehaviorTree.CPP). Instead of pre-calculating a global exploration path, the robot continuously adapts to the environment in real-time.

• The Exploration Loop: The system cyclically fetches the latest frontier points and dynamic maps from the blackboard, applies a closest-frontier policy to select the optimal target, and dispatches the goal to the easynav navigation stack. If a goal fails or is reached, the tree reacts instantly to select the next best frontier.

• Mission Completion: To determine when to stop, the behavior relies on the IsExploredModel, a custom-trained CNN (Convolutional Neural Network) that visually classifies the global map to decide if the area is fully explored, triggering a clean mission termination.

Note: For a detailed technical breakdown of the specific BT action and condition nodes (e.g., ChooseFrontierGoal, GoToPose), please refer to the easynav_multirobot_exploration package README.

Simulation Environments

Collection of Gazebo environments used for multi-robot validation:

• Hospital: The largest and most complex environment for stress-testing path planning.
• Maze: Designed for testing wall-following and frontier logic in tight spaces.
• Maze (Small version): Fast-iteration environment for quick tests.

TO-DO

• Goals blacklist.

• Reactive goals.

• Reinforcement Learning for exploration behaviour.

• Automatic multiplexor map coordination.

• Common map global topic?