A main challenge, in the realm of vehicle guidance systems, consists in the on-line computation of accurate optimal trajectories. In particular, for high-bandwidth plants such as helicopter Unmanned Aerial Vehicles (UAVs), stringent real-time timing constraints may often need to be met. Hence it is the purpose of this paper to present such a novel trajectory planner framework, anchored in the combined paradigms of differential flatness and neural networks, and allowing for computationally tractable optimal control problems. We conclude by presenting simulation examples, for the case of a helicopter UAV in autorotation, that demonstrates the applicability of our proposed approach.