Meeting Abstract
Compared to even the best engineered systems, insects control their flight faster, more robustly, and with greater energetic efficiency. In contrast to engineered systems, recent research suggests insect wing flapping serves the dual roles of both actuation and sensing. Sensory information is acquired by mechanoreceptors that are present on the wings of all insect taxa. These receptors respond to wing strains that result from the combination of flapping motions and body rotations. How the ensemble of wing strain sensors relate local strain information to a control decision remains an open question. We hypothesize that, to control their flight, insects can more robustly determine inertial rotations if those rotational velocities are classified into distinct regimes. To test this hypothesis, we use experimentally determined neural encoding properties of wing mechanoreceptors along with a sparse sensor placement algorithm, and the structural simulation of a model insect wing. We provide our model wing with an array of 25 strain sensors. We can then classify the perturbation to one out of 5 distinct rotational velocities about the yaw axis: [0,5,10,15,20] rad/s. The firing rate of each sensor is weighted and the velocity is determined by linear discriminant analysis. In our simulation, the wing undergoes smooth angular velocity perturbations during 3 seconds, while the combined array of sensors selects one out of the 5 discrete angular velocities at 1 kHz frequency. The model predicts that angular velocity can be discriminated with a mean error of 2.9 rad/s. We suggest that, by sacrificing sensitivity to small velocity changes, the insect can reject sensor uncertainty, thereby increasing the robustness of control output.
