In this paper, we propose integrating a learning-based online estimation with a hybrid control system that combines Model Predictive Control (MPC) and an Adaptive Proportional Controller (APC). This integration aims to improve the reliability and performance of autonomous flight control for unmanned helicopters. We use Subspace and Prediction Error Minimization methods to estimate discrete models of pitch, roll, yaw, and heave motions. The hybrid control system leverages MPC for robustness and performance, while APC adjusts control gains based on the squared error between the target attitude and the predicted output from the online estimator. The online estimator, implemented with a Radial Basis Function Neural Network, uses a weighting matrix to reduce measurement errors and disturbances. Simulations in turbulent conditions show that the proposed hybrid system outperforms standalone MPC in both performance and robustness.