Development of a digital twin for Crazyflie drone swarm testing

Abstract

This thesis explores the development and validation of a game engine-based digital twin system for Crazyflie drone swarms. The limitations of the Crazyflie platform, such as short battery life, dependence on external localisation systems like VICON, and limited onboard sensing, shaped the design requirements of the system. The proposed digital twin was implemented in mixed reality. It allows real-time synchronisation between physical and virtual Crazyflies, simulates perception through virtual LiDAR sensors and cameras, and enables hybrid swarm operations combining real and virtual agents in a shared virtual environment. Unity was selected as the development platform because of its balance of rendering performance, moderate hardware requirements, official support of ROS2, and abundance of learning resources. To extend the capabilities of Crazyflie swarm for research and testing, four features were developed, including dynamic obstacle simulation, integration of virtual sensors and cameras, scalable swarm size through virtual Crazyflies, and autonomous replacement of Crazyflies to extend experiment duration. Experimental scenarios validated its effectiveness for coordinated multi-agent tasks, vision coverage, realistic virtual sensing as well as reduced hardware dependency and risk of hardware damage. This work shows that a game engine–based digital twin can significantly improve the flexibility, scalability, and safety of swarm robotics research while lowering hardware costs and physical space requirements. Future work could enhance model fidelity using AI and extend the system for remote or cloud-based swarm experimentation.