움직임 제약조건과 측정치가 부족한 상황에서의 경사하강법 기반 비순환 최소 퍼시스턴스 그래프를 가진 편대 제어

Abstract

The study focuses on distributed formation control, a field actively explored in multi-object cooperative control. This control method establishes formations in a global coordinate system based solely on information from each agent's local coordinate system. Particularly in formations where the sensing and actuation topology follow a directed graph, the acyclic persistent graph is widely employed, ensuring convergence through various formation control theories such as gradient descent techniques.

Nevertheless, challenges arise when forming a formation with a ground-dependent wheel mobile robot and a drone with an additional degree of freedom perpendicular to the ground. The gradient descent technique exhibits limitations, generating control inputs impractical for the wheel mobile robot. Furthermore, existing formation controllers pose difficulties when applied to agents with non-holonomic constraints, assuming a single integrator model for ease of deployment. Geometric constraints may also emerge based on the attached sensor type, making convergence challenging. Additionally, if complete sensing information required for the gradient descent controller is unavailable, an estimation algorithm must be devised.

This paper addresses gradient descent-based triangle formation control in scenarios involving heterogeneous robots with motion and visibility constraints and insufficient sensing information. The absence of sensing information regarding the locations of other agents is addressed through a distance-based gradient descent adaptation law. Even in cases where a basic gradient descent formation controller fails to achieve the control goal without violating constraints, assurance of convergence is provided by introducing a mode algorithm. This algorithm effectively partitions and applies inputs generated from information obtained from each agent so as not to violate constraints and achieve control goals. The proposed algorithm's efficacy was validated through simulations and experiments.