An Application of Inverse Reinforcement Learning to Estimate Interference in Drone Swarms

Area of Science:

• Robotics and Autonomous Systems
• Artificial Intelligence
• Computational Science

Background:

• Drones have increasing capabilities but limited autonomy for complex missions, leading to vulnerabilities.
• Unanticipated interference significantly impacts drone swarm performance and adaptability in dynamic environments.

Purpose of the Study:

• To develop a computational framework for inferring drone swarm intent from movement data.
• To quantify and analyze interference phenomena in drone operations.

Main Methods:

• Utilized machine learning, including deep learning, to infer interference from movement predictability.
• Employed double transition models and inverse reinforcement learning to derive reward distributions.
• Computed entropy and interference across diverse drone scenarios with varying strategies.

Main Results:

• Drone swarms experienced increased interference, performance, and entropy with greater heterogeneity.
• The direction of interference (positive/negative) was more sensitive to strategy combinations than swarm homogeneity.

Conclusions:

• The framework effectively deduces drone swarm intent and quantifies interference.
• Understanding interference dynamics is crucial for enhancing drone swarm autonomy and mission success.