Multi-Agent Deep Reinforcement Learning Based on Soft Actor-Critic for Self-Collaborating UAVs in a Swarm

This paper presents a novel implementation of a decentralized Multi-Agent Deep Reinforcement Learning (MADRL) framework based on Soft Actor-Critic (SAC) for self-collaborating UAV swarms. The proposed approach enables each UAV to maintain formation, avoid obstacles, and collaborate effectively without the need for explicit communication. By leveraging local observations and learned policies, each UAV independently adapts to dynamic and unpredictable conditions in real-time, making the system robust to environmental variability and scalable to large swarms. The paper defines key SAC equations and parameters tailored for the proposed method, demonstrating its potential in scenarios such as disaster management, and search and rescue missions. A comparative analysis with other MADRL methods highlights the advantages of SAC, explaining the rationale behind its selection.