📌 Safe Learning for Multi-Robot Mapless Exploration

This is the supplementary material for the paper Safe Learning for Multi-Robot Mapless Exploration.

This Python project uses Deep Reinforcement Learning (DRL) to train a TurtleBot3 Waffle Pi for autonomous navigation without relying on pre-built maps. The training enables the robot to perform mapless exploration and avoid obstacles in simulated dynamic environments.

Built within the Robot Operating System (ROS) framework, this implementation leverages Gazebo for realistic simulation and PyTorch for DRL model development.

---

🧠 Exploration Demo

         

---

📚 Citation

If you find this work useful in your research, please consider citing:

@article{liu2025safe,
  title={Safe learning for multi-robot mapless exploration},
  author={Liu, Wenxing and Niu, Hanlin and Wu, Kefan and Pan, Wei and Ji, Ze and Skilton, Robert},
  journal={IEEE Transactions on Vehicular Technology},
  year={2025},
  publisher={IEEE}
}

---

📖 Getting Started

These instructions will help you get the project running on your local machine for development and testing.

---

📋 Prerequisites

Before starting, ensure you have the following installed:

• Ubuntu 20.04
• ROS Noetic
• Python 3.8
• PyTorch 1.10.0
  You can install it using Anaconda to manage a virtual environment or python virtual environment.
• NumPy 1.24.4

---

📦 Installation

Install TurtleBot3 and dependencies:

sudo apt-get update
sudo apt-get upgrade

cd ~/catkin_ws/src/
git clone https://github.com/ROBOTIS-GIT/turtlebot3.git
git clone https://github.com/ROBOTIS-GIT/turtlebot3_msgs.git
git clone https://github.com/ROBOTIS-GIT/turtlebot3_simulations.git
git clone https://github.com/ROBOTIS-GIT/turtlebot3_machine_learning.git
git clone https://github.com/ros-perception/laser_filters.git

cd ~/catkin_ws
catkin_make

---

📥 Clone this Repository

cd ~/catkin_ws/src/
git clone https://github.com/ukaea/SLMRME.git

---

🔧 Modify the TurtleBot3 Laser Scanner Resolution

The default TurtleBot3 Waffle Pi LiDAR sensor publishes 360 rays. For this project, we reduce it to 24 rays to simplify the DRL input and speed up simulation.

📁 File to Modify:

~/catkin_ws/src/turtlebot3/turtlebot3_description/urdf/turtlebot3_waffle_pi.gazebo.xacro

📝 What to Change:

Find this line:

<samples>360</samples>

And change it to:

<samples>24</samples>

🔧 Or Use This Command:

You can automate this edit by running:

sed -i 's/<samples>360<\/samples>/<samples>24<\/samples>/g' ~/catkin_ws/src/turtlebot3/turtlebot3_description/urdf/turtlebot3_waffle_pi.gazebo.xacro

🔄 Rebuild the Workspace:

After making the change, rebuild your workspace:

cd ~/catkin_ws
catkin_make
source devel/setup.bash

Now, your simulated LiDAR scanner will publish 24 rays instead of 360.

---

🚀 Running the Simulation

1️⃣ Start Gazebo World

When we say "starting a Gazebo world", it means launching a simulated environment in Gazebo, which is a powerful 3D robotics simulator integrated with ROS. A Gazebo world is essentially a virtual space with models, obstacles, and physics where your robot can navigate, interact, and be tested safely before deploying in the real world.

What happens when you start a Gazebo world:

• Loads a virtual environment (walls, obstacles, floor)
• Spawns your TurtleBot3 robot model into the environment
• Starts the ROS nodes that publish data like LaserScan, Odometry, and TFs
• Initializes physics simulation for realistic robot behavior

Learn more about Gazebo here: Gazebo: Simulating Robots

Launch the simulation:

roslaunch tvt20_bot single_bot.launch

2️⃣ In a New Terminal — Start DRL Training

roslaunch tvt20_bot single_tvt_torch.launch

🔄 Purpose of the Launch Files

• single_bot.launch 🔺 Starts the robot and Gazebo world

  • Loads the robot model
  • Sets up the simulation environment
  • Initializes ROS nodes and Gazebo physics

• single_tvt_torch.launch 🔺 Starts the Deep Reinforcement Learning training

  • Launches the DRL Python training scripts
  • Subscribes to topics like /scan and /odom
  • Begins training using the simulation data and environment feedback

In short:

• single_bot.launch = Starts the robot + simulated world
• single_tvt_torch.launch = Starts the DRL training process (that uses the simulation)

---

📚 Assumed Background Knowledge

To effectively use this project, you should be familiar with:

• Building and running ROS workspaces
• Gazebo simulator
• Basic concepts of DRL
• Linux terminal operations
• Python and PyTorch basics

If you're new to ROS, a good place to start is the official ROS tutorials.