Simulation for Multi-Agent Autonomy: LTA UAVs in Adversarial Environments

In this paper we present a simulation focused on multi-agent autonomy for lighter-than-air unmanned aerial vehicles in adversarial environments. While autonomous agents have seen increased use in performing real-world tasks, the simulation software for testing these agents has lagged behind. One major difficulty in testing algorithms for multi-agent autonomy is inability to bring software and hardware directly from simulation to the real world. Realistic environments, physical simulators, hardware and software integration, and communication networks are a few of the many requirements for this type of simulation. Originally inspired by the collegiate competition Defend the Republic, this simulation provides users with an interface between Unreal Engine 5 and Python via a TCP connection that allows for software-in-the-loop testing. Hardware-in-the-loop testing will be provided in future updates. Unreal provides photorealistic graphics, in-built physics, and simulated sensors. Data from the environment is sent to Python for processing and control logic and then returned to Unreal so the agent positions can be updated. The developed architecture is intentionally mirrored with the real-world systems used in competition by Indiana University to facilitate transfer between simulation and real-world.