A method to implement and to evaluate a learning-based Piloting Autonomous System for UAS

The growing public interest for Unmanned Air Systems (UAS) applications has stimulated the debate over the integration of this kind of aircraft into the civil aviation system. However, the concept of not having a human pilot inside the aircraft presents uncertainties that may impede the creation of proper regulation. Having safety as the main concern for civil aviation, one important principle of aviation to be addressed in an UAS is collision avoidance, a traditionally pilot-dependent functionality. Therefore, we propose a method for implementing a learning-based Piloting Autonomous System. The proposed approach utilizes the concept of `Learning from Demonstration´ in order to define a behavior for the autonomous aircraft based on the maneuvers executed by a human pilot. By doing so, we expect the Piloting Autonomous System to be able to compensate for the lack of a human pilot in the aircraft. Therefore, the proposed approach would represent a possible implementation of an autonomous UAS that presents the same safety levels observed in (human-based) civil aviation. Additionally, we identify metrics that can be used to select a suitable learning-based method and to compare its performance to those observed in manned aircraft.