Daniel, Thomas L.Bustamante, Jorge2021-10-292021-10-292021-10-292021Bustamante_washington_0250E_23470.pdfhttp://hdl.handle.net/1773/47945Thesis (Ph.D.)--University of Washington, 2021Research on insect flight control has focused primarily on the role of wings. Yet airframe de- formations via abdominal deflections during flight may potentially influence the dynamics of flight and play a significant role in control and management of energy resources. To address these general issues, I use a combination of a Model Predictive Control (MPC)-inspired computational inertial dynamics model, free flight experiments in the hawkmoth, Manduca sexta, and–in collaboration with another research group–bio-inspired machine learning methods. The 2D inertial dynamics model simulated underactuated and fully actuated models tracking a vertically oscillating flower. Flight performance was evaluated with two metrics: tracking error and cost of transport. The models suggest that fully actuated simulations minimized the tracking error and cost of transport. Moreover nearly eliminating the role of the abdomen by reducing its mass significantly worsened both flight performance metrics. Additionally, I tested the effect of restricted abdomen movement on free flight in live hawkmoths by fixing a carbon fiber rod over the thoracic-abdomen joint. Hawkmoths with a restricted abdomen flew less frequently compared to the sham treatment moths. Furthermore those moths which flew, performed worse than sham treatment moths in various flight performance metrics. I also explored inertial-elastic, and morphological modifications of the dynamics model to examine the trade-offs between flower tracking accuracy and flight performance for a range of size, shapes, abdominal masses, and thoracic-abdomen stiffness values. I found that increasing abdominal mass reduced tracking error across size ranges while also increasing the mechanical work and cost of transport. Moreover, increasing petiole length reduced tracking error across size ranges while also increasing the mechanical work and cost of transport. Lastly, in a collaborative study, we used deep neural networks (DNNs) trained by MPC data sets to develop an efficient method for solving inverse flight control problems. Furthermore, these DNNs were pruned to determine how sparse a net- work can be and still yield successful flight. We find that the network can be pruned to ∼7% of the original networks weights and still fly successfully. Motions of non-aerodynamic structures, found in all flying animals, and the use of pruning DNNs can inform the development of multi-actuated micro air vehicles as well as determine the role the abdomen plays in insect flight control and energetics.application/pdfen-USnonebiomechanicsflight controlinsect flightmodel predictive controlBiomechanicsBiologyThesis