Document Type

Presentation

Date of Original Version

3-29-2025

Abstract

Multi-agent systems (MASs) have become a powerful approach for solving complex real-world problems due to their efficiency, scalability, and adaptability. MASs consist of multiple agents that collectively perform tasks, often requiring precise coordination, collision avoidance, and efficient navigation. Formation control is essential in such systems, regulating agents' positions, velocities, and orientations to maintain specific patterns. Formation control methods are categorized into three approaches: (1) Behavioral, where agents follow predefined behaviors derived from subsolutions; (2) Leader-following, where agents maintain a configuration relative to designated leaders but are vulnerable to leader faults; and (3) Virtual structure, where agents follow a virtual leader, enhancing system robustness. These approaches operate under centralized or decentralized control. Most leader-following techniques assume linear time-invariant leader dynamics without inputs, limiting their ability to handle complex formation tracking tasks. Additionally, challenges related to dynamic variations in inertia remain unresolved. To address these issues, we propose a generalized leader dynamics model with a virtual leader subject to bounded time-varying external inputs. We develop a novel cooperative deterministic learning-based adaptive formation control scheme for nonlinear mechanical systems with uncertain dynamics. Our framework consists of two layers: (1) a cooperative nonlinear estimation protocol to estimate leader 24 states and (2) a cooperative deterministic learningbased formation control protocol using radial basis function neural networks (RBF NNs) for formation tracking and accurate learning of nonlinear dynamics. Numerical simulations confirm that our approach ensures formation tracking while achieving cooperative and accurate learning of system dynamics with guaranteed convergence and learning consensus.

Download

Share

COinS