System Identification for Fully-Actuated UAV Control Allocation

Control allocation in multirotor unmanned aerial vehicles (UAVs) converts control signals to commands for individual actuators. Errors may be introduced into the system via inaccurate or insufficient control allocation due to aerodynamic effects, saturation, or hardware asymmetries. This paper presents a methodology to experimentally estimate such control allocation errors for a UAV in free-flight using motion-capture. The methodology is tested with a fully-actuated octocopter, and steady-state allocation errors are identified and investigated. A generalized experimental approach to deriving control allocation is then introduced. The approach uses system identification to model-fit the control allocator of a UAV from free-flight data. The approach was tested and reduced allocation error by 22% for the fully-actuated octocopter. Finally, a matrix representation for comparing control allocators is introduced and used to identify allocation error causes.