Transient-Buffered Allocation-Geometry Adaptation for Fully Actuated Multirotor UAVs Under Center-of-Mass Variations

Fully actuated multirotor UAVs (FA-mUAVs) have attracted attention for aerial physical interaction because they generate translational forces without tilting the vehicle, thereby supporting interaction precision and payload-transportation stability. However, this capability relies on an accurate center-of-mass (CoM)-dependent thrust-to-wrench allocation geometry. When the CoM shifts due to payload variation or physical interaction, the nominal allocation geometry produces incorrect moment arms and recurrent unintended torques. Existing robust and robust-adaptive approaches compensate for these effects through the input channel while leaving the allocation geometry unchanged. Conversely, allocation-level CoM correction addresses the geometric source of the mismatch but does not by itself provide transient robustness during estimator convergence or against non-CoM disturbances. This paper proposes a transient-buffered allocation-geometry adaptation framework that addresses these two limitations within a single architecture. The framework combines an input-channel robust layer with allocation-level CoM adaptation: the robust layer buffers instantaneous mismatch and residual disturbances, while the adaptation layer processes the filtered compensation signal generated by the robust layer through the CoM-dependent regression to estimate the CoM and update the allocator, without assuming a unique separation of CoM-induced and non-CoM components. Stability analysis establishes boundedness and phase-dependent convergence of the CoM-estimation error. Real-flight experiments show that adding the CoM-estimation/allocation-update path to a retained disturbance-observer (DOB) robust baseline improves attitude regulation and reduces rotational compensation demand.