Efficient and resilient micro air vehicle flapping wing gait evolution for hover and trajectory control
Engineering applications of artificial intelligence
This paper deploys a recently proposed, biologically inspired, on-line, search-based optimization technique called Selective Evolutionary Generation Systems (SEGS) for control purposes; here, to evolve Micro Air Vehicle (MAV) flapping wing gaits in changing flight conditions to maintain hovering flight and track trajectories in unsteady airflow. The SEGS technique has several advantages, including: (1) search-efficiency, by optimally trading off prior search space information for search effort
... avings as quickly as possible in dynamic environments; (2) model-independence, as in biology, avoiding biases induced by built-in models rendered incorrect by environment changes; and (3) resilience, through sufficiency for stochastic behavior that is itself sufficient for responsiveness to search-objective variations caused by environment fluctuations. This work presents the first approach that can simultaneously evolve optimal MAV flapping wing gaits efficiently and resiliently, adapt on-line, and, via model-independence, allow feedback from either experimental sensors or alternate external models (affording control versatility for hover or forward flight, unsteady or quasi-steady aerodynamics, and any dynamics or wing kinematics). Performance benchmarks are also provided. Because the (1+1)-Evolution Strategy (ES) and the Canonical Genetic Algorithm with Fitness Proportional Selection (CGAFPS) are two SEGS special extreme cases, an additional comparison showcases SEGS possession of both (1+1)-ES computational speed and CGAFPS resilience. Highlights 1. A bioinspired, search-efficient, tunable optimization scheme is adapted for control. 2. Micro Air Vehicle (MAV) flapping wing gaits are evolved on-line, model-independently. 3. Scheme properties are benchmarked in a case study of evolution for MAV hover control. 4. Scheme speed and responsiveness compare favorably to related evolutionary methods. 5. A second study attains MAV trajectory control in unsteady flow with little computing.