Onboard/Offboard Sensor Fusion for High-Fidelity Flapping-Wing Robot Flight Data

摘要

for a novel and complex type of vehicle, that is not known in great depth.However relatively limited work has been done in this field so far, largely due to the difficulty of obtaining suitable data.On the other hand, many low-order dynamic and aerodynamic models currently available have not been validated with free-flight data, and particularly not in different flight regimes [26,27,28,29].Thus, a significant limitation in the analysis, modelling and design of flapping-wing vehicles, is the limited free-flight data available.Particular challenges include obtaining accurate free-flight data: (i) at the high frequencies relevant for flapping flight, (ii) in different flight regimes and (iii) during maneuvers.Such data would for instance support the development of more realistic models, possibly covering different flight regimes, a better understanding of maneuvering flight, and a better evaluation of free flight as opposed to the tethered wind tunnel case.This in turn would support improved understanding and further development of flapping-wing vehicles.This paper proposes a sensor fusion-based method for the acquisition of accurate free-flight data for flapping-wing vehicles.Fusion of on-board inertial measurement unit (IMU) data and off-board optical tracking data is suggested as an approach to obtain high quality, more reliable measurements, that are also accurate at high frequencies and can be used for in-depth analysis of time-resolved flapping effects in free flight and during maneuvers.External optical tracking systems [8,30,31], such as are often used for FWMAVs, provide accurate position measurements, but typically have a relatively low bandwidth if the intention is to consider processes happening during each flap cycle.Additionally, linear and angular velocities and accelerations have to be computed via numerical differentiation, which introduces considerable noise [16].It can also be expected that as platforms become increasingly small, attitude data obtained from tracking alone will decrease in quality due to the limited resolution.By contrast, IMU sensors have a high bandwidth, which allows for high-frequency measurements.Accelerations and angular velocities are measured directly, however both are affected by integration drift -when used to determine attitude or velocity -and high noise levels.Hence, it is suggested that combining the approaches will yield more accurate and reliable data than either of the two approaches alone, and, in particular, data providing insight into what is happening during each flap cycle.The forces computed from the obtained data are compared with wind tunnel measurements, to provide an initial idea of their accuracy and reliability.Additionally, implications of the suggested data processing approach for time-resolved modelling are discussed.We argue that our approach is advantageous for analysis of time-resolved aerodynamics and dynamics, giving results that are qualitatively comparable to wind tunnel ones, but more realistic since representing free flight, and providing the opportunity of considering dynamic maneuvers.The resulting data allow for the forces acting on the FWMAV in free flight (including maneuvers) to be reconstructed with high accuracy and resolution, at smaller time scales than previously considered.Meaningful information is obtained at frequencies up to five times the flapping frequency of the test vehicle, 2 of 21