Challenges for Safe Autonomous Flight

Abstract

Autonomy holds a great promise by improving the applications, safety, and efficiency of flight. If little operator input is necessary, unmanned rotorcraft have a wide range of applications ranging from cargo delivery to inspection. Currently unmanned rotorcraft are underutilized because they either have to fly on preplanned missions at high altitude or require careful teleoperation. A capable autonomous rotorcraft will have to react quickly to previously unknown obstacles, land at unprepared sites, and fly with semantic information to enable long-term autonomy in cluttered environments. In this talk we give an overview of how autonomy systems are currently designed and how pushing the performance and safety of these systems challenges current verification paradigms. In particular we will address how a supervisory layer in the motion planning system can improve safety, a sensor steering system enables us to optimize coverage for safe trajectories, and how semantic information can help us guide the rotorcraft.