Data-driven flocking behavior of multi-agent systems under uncertain environments

Abstract

In numerous practical application scenarios, the collaborative control tasks of multi-agent systems are challenged by the fact that the dynamic characteristics and environmental conditions of agents (such as robots, drones, and autonomous vehicles, etc.) are often unknown. In this paper, an approach based on data-driven learning control barrier functions is proposed to investigate the flocking control issue of multi-agent systems with unknown dynamics under uncertain environments. The verification of the flocking behavior in multi-agent systems is facilitated by flocking error system, upon which a data-driven flocking controller employing pseudo partial derivatives is constructed. Secondly, a data-driven learning control barrier functions scheme based on the idea of Gaussian process is established for multi-agent systems under uncertain environment, which can ensure the safety of agents and eliminate the dependence of traditional control barrier functions method on the system model. Moreover, two zeroing control barrier functions are provided to ensure the safety and low uncertainty of the agents, respectively, so as to achieve system convergence. Finally, the effectiveness of theoretical results is verified through numerical simulation and experiments.