Robust formation control of heterogeneous UAV-USV multi-agent systems: An actor-critic enhanced hyperbolic sliding...

This paper investigates the robust formation control problem for heterogeneous Unmanned Aerial Vehicle (UAV) and Unmanned Surface Vehicle (USV) multi-agent systems in mission-critical maritime scenarios (e.g., search and rescue) subject to complex environmental disturbances and strict communication constraints. A synergistic control architecture integrating an Actor-Critic Reinforcement Learning (RL) algorithm, a Hyperbolic Sliding Mode Control (HSMC) law, and a Dynamic Event-Triggered Mechanism (DETM) is proposed. To resolve the inherent theoretical conflict between discontinuous sliding mode switching and Zeno-free event-triggered updates, a continuous hyperbolic tangent approximation is rigorously introduced. Consequently, the closed-loop coupled continuous-discrete dynamics are proven to achieve Uniformly Ultimately Bounded (UUB) stability, ensuring that all tracking errors converge to a compact Quasi-Sliding Mode Band. Furthermore, a novel gain-threshold co-design is developed, utilizing a low-pass filter to harmonize the RL agent's trial-and-error exploration with the DETM's communication sparsification. Instead of computationally heavy deep networks, the proposed Actor-Critic RL employs lightweight linear basis functions, yielding constant per-agent adaptation overhead and an overall online computational complexity that scales approximately as O(N) with the number of follower USVs. Extensive simulations, explicitly incorporating realistic physical constraints such as stochastic sensor measurement noise, actuator dynamics and saturation, and network packet losses, validate the practical engineering value of the proposed architecture. Ablation studies demonstrate that the proposed framework achieves superior transient tracking performance and robustness compared to optimally tuned baselines while significantly reducing communication transmission loads by over 50%.