MODIFLY: A Scalable End-to-end Multi-Agent Simulation for Unmanned Aerial Vehicles
Keywords: Multi-agent Systems, Unammaed Aerial Vehicles (UAVs), Reinforcement Learning
Abstract: Multi-agent unmanned aerial vehicle (UAV) systems have emerged as a promising solution for complex applications such as industrial automation, surveillance, and disaster response. However, the application of multi-agent UAV coordination remains challenging due to the lack of consideration of real-world constraints such as communication link degradation, scalability issues, and the need for realistic training environments. Existing simulation platforms often lack the fidelity and flexibility required to bridge the gap between simulation and deployment. To address these limitations, we propose MODIFLY, a scalable, cross-platform, end-to-end simulation platform tailored for multi-agent UAV control. MODIFLY introduces dynamic communication modeling, including link degradation, to accurately simulate real-world UAV operations. It supports distributed execution across multiple UAVs, seamless coordination, real-time monitoring, and user input capture. MODIFLY uniquely integrates real drones with virtual environments, allowing UAVs to interact with simulated obstacles and peer ones for hybrid reality testing. Additionally, the platform is designed to facilitate reinforcement learning (RL) research by providing compatibility with popular libraries like OpenAI Gym and PettingZoo, supporting both single-agent and multi-agent RL environments. MODIFLY offers an intuitive interface for real-time parameter tuning and performance analysis, making it a versatile tool for researchers and practitioners to develop and validate UAV coordination strategies under realistic conditions.
Submission Number: 18
Loading