Statistical verification for complex controllers with applications to unmanned aircraft

Research in advanced control has had limited impact on real-world applications. We believe a principal cause is that the computational and performance properties of complex controllers cannot be feasibly analyzed using traditional (i.e., deterministic) verification and validation techniques. Statistical methods present an alternative. We discuss how extensions of statistical learning theory can provide rigorous estimates of reliability. Applications to computational tractability (e.g., under what conditions can we be assured that an iterative control calculation will terminate within an allocated time interval?) and critical performance aspects (e.g., under what conditions can high-performance maneuvers be successfully executed?) are presented. The applications pertain to new unmanned aerial vehicles (UAVs) that are designed for urban environments.