Intelligent swarm UAV navigation system with obstacles avoidance and multi-target tracking capability

With the advancement of technology, the use of unmanned aerial vehicles (UAVs) is increasing in both military and civilian sectors. Particularly, UAV swarms offer an effective solution in emergencies where rapid decision-making and coordinated action are critical. Research efforts have focused on enhancing the autonomous and coordinated behaviours of UAV swarms to serve as a beneficial tool for assisting society in emergency situations. Swarm intelligence, a branch of artificial intelligence, simulates the interactions and self-organization of multiple entities, playing a significant role in optimization, engineering, and multi-UAV collaboration. Swarm intelligence algorithms provide substantial academic and practical benefits in critical tasks such as task assignment, route planning, collision avoidance, and formation reconfiguration for multi-UAV missions. The use of deep reinforcement learning (DRL) technology is becoming increasingly important for optimizing UAV swarm navigation. DRL enables UAVs to continuously interact with their environments to acquire the best navigation techniques, allowing them to collaborate to achieve group objectives. The multi-agent deep reinforcement learning (MADRL) model can be applied to various scenarios, including multi-UAV control, multiplayer games, and multi-robot control. This research develops a decentralized intelligent UAV swarm navigation system capable of real-time collision avoidance and multi-target tracking in unknown environments. At the core of this system is the Multi-Agent Actor-Critic with Reciprocity (MAAC-R) algorithm, which optimizes both individual and collective swarm behaviour through a reciprocal reward mechanism. This approach allows UAVs to dynamically adjust their strategies based on their impact on neighbouring agents, promoting better coordination without the need for global information, which is a crucial feature in dynamic and unpredictable environments. When integrated with MSE-based optimization, MAAC-R outperforms other methods such as MAAC-G and simple MAAC, offering greater flexibility, scalability, and adaptability. These characteristics make it particularly suitable for complex scenarios, such as disaster response and search-and-rescue operations. Additionally, the system incorporates advanced LIDAR-based collision avoidance capabilities, enabling UAVs to detect obstacles in real-time and adjust their flight paths accordingly. All these innovations position MAAC-R(MSE) as a comprehensive and powerful solution for tasks requiring efficient multi-target tracking and safe operation in challenging, obstacle-filled environments.