Identification, control and visually-guided behavior for a model helicopter

Research on unmanned aerial vehicles is motivated by applications where human intervention is impossible, risky or expensive e.g. hazardous material recovery, traffic monitoring, disaster relief support, military operations etc. Due to its vertical take-off, landing and hover capabilities, a helicopter is an attractive platform for such applications. There are significant challenges to building an autonomous robotic helicopter - these span the areas of system identification, low-level control, state estimation, and planning. Towards the goal of fully-autonomous helicopters this thesis makes the following contributions. A continuous-discrete extended Kalman filter has been developed that combines inertial data with GPS and compass data to provide estimates of the 6DOF state of the helicopter. Using this filter a model for the helicopter has been identified based on frequency response techniques. The model has been validated in flight tests on a small helicopter testbed (1.6 m rotor diameter) at speeds upto 5 m/s. Based on evidence from this model a decoupled low-level controller has been developed which is embedded in a control architecture suitable for visually-guided navigation. As a novel application, we show how such a controller can be used to perform trajectory following on the helicopter where the desired trajectories are typical spacecraft landing trajectories, and the only controls available are thrusters. This in effect, produces a low-cost testbed for testing spacecraft landing and hazard avoidance on a planetary surface. Finally, we develop and extensively experimentally characterize algorithms for vision-based autonomous landing, object tracking, and sensor deployment.