Safe Optimal Control of Cooperative Multi-Agent Systems
- Publication Type:
- Thesis
- Issue Date:
- 2025
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
The cooperative control of multi-agent systems (MASs) has garnered significant attention due to its broad range of applications, including environmental exploration, area coverage, and more. Two fundamental requirements for MASs are stability, which ensures that the system states converge to desired objectives, and safety, which guarantees that agents operate within safe regions. To meet these requirements, control Lyapunov functions (CLFs) are employed to achieve stability, while control barrier functions (CBFs) are used to enforce safety constraints. These functions are integrated into a unified quadratic programming (QP) framework, where control inputs are computed by solving constrained optimization problems.
In practical applications, external disturbances and modeling uncertainties are often inevitable, thereby necessitating the development of robust control strategies. Within the proposed QP control framework, such uncertainties typically appear as unknown components in the constraint formulations. One conventional approach to address this issue is the use of model-based robustness techniques, which rely on bounds of uncertainties. Alternatively, modern data-driven approaches, such as Gaussian process (GP) modeling, offer possible solution by providing probabilistic approximations of the unknown functions, including mean and variances predictions. This probabilistic characterization enables an estimation of uncertainty, enhancing the robustness and overall reliability of the control framework.
Moreover, MASs frequently operate in resource-constrained environments, where control performance is of high importance. In such scenarios, system behavior is often influenced by hyperparameters whose effects on performance are complex and difficult to model analytically. To systematically improve long-term control performance, this thesis formulates this problem as a black-box optimization task and proposes a constrained Bayesian optimization (CBO) algorithm, which efficiently explores the hyperparameter space to identify configurations that optimize system performance while ensuring stability and safety.
Building upon the above statement, this thesis investigates the cooperative control of MASs from the following main perspectives. First, establish a constrained QP control framework that ensures both stability and safety for decentralized MASs; Second, enhance robustness to uncertainties by dealing with unknown disturbances, thus enabling the control system to adapt in real time to uncertain dynamics; Third, optimize control performance through a CBO algorithm that explicitly considers both safety and effectiveness under varying hyperparameter settings. The proposed methodologies are validated through extensive simulations and case studies, demonstrating their effectiveness in achieving stable, safe, robust, and high-performing cooperative control in MASs.
Please use this identifier to cite or link to this item: