Text this: Residual reinforcement learning for disturbance-resilient control under modeling uncertainties.