Meta-Learning-Based Robust Adaptive Flight Control Under Uncertain Wind Conditions

Realtime model learning proves challenging for complex dynamical systems, as drones flying in variable wind conditions. Machine learning technique as deep neural networks have high representation power but is often too to update onboard. On the other hand, adaptive control relies on simple parameter models can update as fast as the feedback control loop. We an online composite adaptation method that treats outputs from a deep network as a set of basis functions capable of representing different conditions. To help with training, meta-learning techniques are used to the network output useful for adaptation. We validate our approach by a drone in an open air wind tunnel under varying wind conditions and challenging trajectories. We compare the result with other adaptive with different basis function sets and show improvement over and prediction errors.