Robotics Control Systems

Heterogeneous Crazyflie + TurtleBot3 Swarm Control System

Project Overview

This project develops a unified control, simulation, and autonomy framework enabling heterogeneous swarm coordination between Crazyflie UAVs and TurtleBot3 ground robots. The system integrates ROS 2, Gazebo, sensor fusion, and reinforcement-learning-based controllers, creating a scalable testbed for swarm robotics research.

Engineering Goals

• Build a ROS 2 native control stack for multi-robot coordination.
• Integrate Crazyflie UAVs and TurtleBot3 platforms into shared formation control behaviors.
• Customized Python/C++ firmware extensions for real-time telemetry, sensing, and distributed control.
• Prototype a PyTorch reinforcement-learning controller to replace classical PID loops.
• Establish a simulation-first workflow in Gazebo to validate multi-agent algorithms before deployment.

System Architecture

1. ROS 2 Control Framework

• Custom ROS 2 nodes for:

  • Formation control
  • Trajectory tracking
  • Multi-robot communication
  • State estimation using LiDAR + IMU

• Centralized communication for synchronized aerial/ground operation.

2. Firmware Development

• Extended Crazyflie firmware in Python/C++ for:

  • High-rate sensor acquisition
  • Distributed telemetry logging
  • Real-time control input processing

3. Gazebo Simulation Environment

• Full simulation of:

  • TurtleBot3 ground robots
  • LiDAR + IMU sensor data
  • ROS 2 multi-agent interactions

• Supports:

  • Hardware-in-the-loop testing
  • Parameter optimization

Machine Learning Integration

Reinforcement Learning Flight Controller

• Built PyTorch RL controller to optimize flight stability and PID behavior.
• Live inference executed directly within ROS 2 nodes.
• Trained using simulated disturbances and noise to improve robustness.

Heterogeneous Swarm Coordination

• Crazyflie drones and TurtleBot3 platforms share:

  • Position and velocity information
  • Formation roles and task assignments
  • Mission-level commands

• Implemented formation shapes:

  • Line
  • V-shape
  • Column
  • Mixed-role leader–follower

• Ensures consistent behavior across:

  • Real hardware
  • Gazebo simulation
  • Hybrid (HIL) setups

Engineering Contributions

• Designed ROS 2 swarm coordination pipeline.
• Developed real-time swarm communication using LiDAR/IMU fusion.
• Implemented distributed control for mixed UAV–ground robot teams.
• Built ML-in-the-loop controllers improving stability over classical PID.
• Created a reproducible, scalable research platform for future autonomous swarm studies.

Key Technologies

ROS 2, Python, C++, Gazebo, Crazyflie Firmware, PyTorch, LiDAR/IMU Sensor Fusion, Reinforcement Learning, Distributed Systems, Autonomous Navigation

Results

• Achieved stable real-time communication between aerial and ground robots.
• Demonstrated coordinated formation behaviors in simulation and hardware.

• Built a platform suitable for:

  • Multi-robot research
  • Controls experiments
  • Swarm autonomy education
  • ML testing in robotics

Future Work

• Deploy RL controller on physical Crazyflie hardware for full closed-loop evaluation.
• Extend swarm to >4 robots with hierarchical control layers.
• Publish the platform as an open-source heterogeneous swarm framework.