Real-Time computation of safe input sequences for unmanned aerial vehicles

A new methodology for incorporating simulation-based reachability analysis in determining the safe set of input signals over a finite time horizon was studied. By analyzing the trajectories obtained using the aforementioned input definition, it appears that sufficient variation in the final state is achieved to cover potential controller input. By representing the future safe input signals compactly in the form of an eight-dimensional ellipsoid and by using machine learning to characterize this ellipsoid, a low-powered embedded system can readily access the output of these extensive offline computations. The methodology enables real-time systems to obtain the coefficients of the safety ellipsoid with little computational effort, which is well within the time required to generate a new control solution.