Improvements to Speed and Efficacy in Non-Stationary Learning in a Flapping-Wing Air Vehicle: Constrained and Unconstrained Flight

Abstract

Small Flapping-Wing Micro-Air Vehicles (FW-MA Vs) may experience wing damage and wear while in service with even small amounts introducing significant deficits in maintaining path control. Previous work employed a custom Evolutionary Algorithm (EA) that adapted wing motion patterns, while in flight and in normal online service, to compensate for wing damage. Although generally successful in finding solutions to this challenging online non-stationary problem, the previous methods would very often require hours of flight time to reach full success and sometime failed altogether in cases of extreme wing damage. This paper details a new approach that reduces the required learning time by an order of magnitude and extends the range of damage over which one can expect suitable performance. A discussion of what changes were made and why they were made will be provided along with extensive simulation results demonstrating the claims of success. The paper will also provide discussion of what additional work is possible now that both speed and efficacy have been sufficiently improved to support practical in-flight learning in real vehicles.