This paper proposes a novel dual-layer learning-based cascade architecture, termed LAMDA-LAMDA, for cooperative formation control of aerial manipulators. The strategy integrates two hierarchical LAMDA controllers: an inner loop that performs velocity-level dynamic compensation and disturbance attenuation, and an outer loop that regulates formation shape and centroid tracking. Unlike conventional model-dependent approaches, the proposed control law does not require explicit knowledge of the aerial manipulator dynamics, which are characterized by strong nonlinear coupling between the hexacopter platform and the onboard manipulator. A Lyapunov-based stability analysis guarantees asymptotic convergence of both velocity and formation errors under bounded uncertainties. The controller is benchmarked against four reference schemes: Kinematic-SMC, SMC with Inverse Dynamics (SMC-ID), SMC-SMC cascade, SMC-LAMDA, and LAMDA-LAMDA cascade, considering abrupt reference changes and severe parametric disturbances affecting inertia, Coriolis, and gravitational terms. Quantitative results show that LAMDA-LAMDA achieves the lowest tracking errors, with average ISE = 0.702 and IAE = 1.652, corresponding to improvements of 35.3% and 32.1% over the best model-based alternative. Additionally, the proposed scheme generates smooth control actions while preserving robustness, highlighting its suitability for cooperative aerial manipulation under dynamic uncertainty.
Andaluz et al. (Sat,) studied this question.