swarm_control

A ROS 2 package for decentralized swarm coordination and control using a finite-time kinematic model and dynamic leader election based on graph connectivity.

This package simulates a multi-robot system (e.g., TurtleBot3s in Gazebo) where each agent interacts only with its neighbors within a vision radius.
The system collectively tracks a reference trajectory in a distributed and leader-adaptive manner.

Package Overview

swarm_control/
├── CMakeLists.txt
├── launch
│   ├── empty_world.launch.py
│   ├── swarm_launch.py
│   └── swarm_mpc_launch.py
├── models
├── package.xml
├── params
│   ├── minimal_bridge.yaml
│   └── swarm_config.yaml
├── rviz
│   └── swarm.rviz
├── swarm_control
│   ├── graph_observer.py
│   ├── kinematic_node.py
│   ├── mpc_controller.py
│   ├── pose_publisher_node.py
│   ├── reference_node.py
│   ├── visualizer_node.py
│   └── utils
│       └── namespace_utils.py
├── urdf
│   ├── common_properties.urdf
│   ├── minimal_urdf.urdf
│   └── turtlebot3_waffle.urdf
├── worlds
│   ├── empty_world.world
│   └── tb3_world.world
└── README.md

swarm_control_msgs/
├── CMakeLists.txt
├── package.xml
├── msg/
│   ├── Info.msg
│   └── RBroadcast.msg
└── README.md

Parameters

Parameter	Node	Description	Default
num_bots	all	Number of robots in the swarm	3
bot_id	kinematic_node	Robot namespace ID (e.g. "bot1")	—
role	kinematic_node	"leader" or "agent"	"agent"
delta_radius	graph_observer	Vision/communication radius [m]	1.5
sampling_freq	reference_node	Sampling Frequency of reference [hz]	2.0
control_freq	mpc_node	Controller Frequency of MPC node [hz]	10.0

Running the Simulation

1. Build

cd ~/ros2_ws/src
git clone https://github.com/BracedHornet186/swarm_simulations.git
export GZ_SIM_RESOURCE_PATH=$(pwd)/swarm_control/models/:$GZ_SIM_RESOURCE_PATH
cd ~/ros2_ws
colcon build --symlink-install
source install/setup.bash

2. Launch Swarm Control Stack

ros2 launch swarm_control swarm_mpc_launch.py num_bots:=3 

This will:

• Spawn 3 robots in Gazebo
• Launch one kinematic_node.py per bot
• Launch one mpc_controller.py per bot
• Start pose_publisher_node.py
• Start graph_observer.py
• Start reference.py

Topic Overview (per bot)

Topic	Type	Publisher	Description
/bot_i/pose	geometry_msgs/PoseStamped	Gazebo	Ground-truth position
/bot_i/delta	geometry_msgs/PointStamped	kinematic_node	Local Δ(t) state
/bot_i/info	swarm_control/Info	graph_observer	Status + role
/r_broadcast	swarm_control/RBroadcast	reference	Leader’s reference broadcast
/reference	geometry_msgs/PointStamped	reference	True reference trajectory
/bot_i/cmd_vel	geometry_msgs/Twist	mpc_node	Control input (velocity)

Algorithmic Flow

1. Graph Construction: graph_observer builds a time-varying adjacency graph based on inter-robot distances.
2. Leader Election: The robot with the highest degree in each connected component becomes leader.
3. Reference Broadcast: Leaders publish the reference r(t) to /r_broadcast.
4. Decentralized Control: Each agent computes Δᵢ = zᵢ − r(t) using neighbor states and performs the finite-time update.
5. Dynamic Reconfiguration: When components merge/split, graph_observer redefines leaders and publishes updated roles.
6. MPC Node: Solves the control problem using differential drive model to publish /bot_i/cmd_vel input.

Extending the Package

• Add a Model Predictive Control (MPC) layer for velocity control using /bot_i/delta as the input reference.
• Include a visualization node publishing visualization_msgs/Marker lines for graph edges in RViz.
• Implement health metrics in Info.msg for real-world diagnostics.
• Integrate rclpy.lifecycle nodes for fault-tolerant reconfiguration.

Dependencies

• ROS 2 Jazzy
• Gazebo Harmonic
• geometry_msgs
• rclpy
• numpy

Author & Maintainers

Developed by Yash Purswani as part of a decentralized swarm control project for ME5253: Network Dynamics and Controls at IIT Madras.