Multi-Agents Cooperative Target Tracking Under Physical Attacks With Environment-Aware Dynamic Constraints

In this work, we consider a cooperative target tracking problem by multi-agents in a 3D space, under disruptions from multiple physical attackers, where the target and physical attackers are intelligent, that is, they adjust velocities based on relative coordinates with agents and agents’ velocities. Due to the presence of physical attackers, some safety and performance constraint requirements cannot be merely constant or time-varying. Specifically, safety constraint requirements on inter-agent distances and performance constraint requirements on formation tracking errors are environment-aware and dynamic, which depend on distances between agents and physical attackers. We propose a neural network-based adaptive cooperative control framework for the agents, which incorporates a universal barrier function to handle safety and performance constraints. We show that formation tracking errors are uniformly ultimately bounded, while all safety and performance constraints are met. A comparative simulation study further illustrates the efficacy of the proposed framework. Note to Practitioners—Cooperative target tracking control for multi-agent systems has gained significant attention due to its real-world applications. In practice, target-tracking tasks are often carried out in complex environments, where multiple physical attackers can fly nearby to disrupt the agents. Moreover, the target and attackers often possess a certain level of intelligence, making the target more difficult to track, complicating an already difficult problem. Thus, most control frameworks in the literature, which assume targets and attackers have constant or time-varying velocities, are not effective to address real-world operational complexities. Additionally, in complex environments, certain constraint requirements need to be environment-aware and dynamic, instead of being mere constants or time-varying. To address these practical challenges, this work proposes a neural network-based adaptive cooperative control framework that integrates a universal barrier function to ensure that all safety and performance constraints are met. Finally, we consider two scenarios in the simulation study to further validate the efficacy of the proposed control framework.