Abstract

This paper presents a unified framework for the safe and optimal control of heterogeneous quadrotor unmanned aerial vehicles (QUAVs) in formation, enabling multitask missions without requiring precise system dynamics. To address partial state observability, a multilayer neural network (MNN) observer is designed to estimate unmeasured states. Reinforcement learning (RL) is employed for optimal control utilizing an MNN ensuring adaptability. Barrier Lyapunov Functions (BLFs) are integrated into the RL framework to enforce safety by maintaining QUAVs within predefined constraints. An enhanced continual learning (ECL) method is proposed to improve the adaptability of MNNs. This method enables effective multitask learning while mitigating catastrophic forgetting. A 3D formation strategy based on spherical coordinates is utilized for leader-follower coordination. This ensures precise and efficient maneuvering. Theoretical analysis establishes system stability. Simulation results validate the proposed framework, demonstrating its effectiveness in achieving safe, optimal, and adaptable formation control for heterogeneous QUAVs in multi tasks missions.

Department(s)

Electrical and Computer Engineering

Second Department

Computer Science

Publication Status

Early Access

Comments

Army Research Office, Grant W911NF-22-2-0185

Keywords and Phrases

Continual Learning; Formation Control; Multi-layer Neural Networks; Observer Design; Optimal Control

International Standard Serial Number (ISSN)

1557-9603; 0018-9251

Document Type

Article - Journal

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2025 Institute of Electrical and Electronics Engineers, All rights reserved.

Publication Date

01 Jan 2025

Download

Full Text Link

Share

 
COinS