06.1 - Flight Dynamics and Control (Control & Modelling)
META-REINFORCEMENT LEARNING-BASED FAULT TOLERANT CONTROL FOR A QUADROTOR WITH A SEVERE LOSS OF ROTOR
S.-H. Kimı, Y. Jungı, Y. Kimı; ıSeoul National University, South Korea
This study proposes a design scheme to obtain an optimal fault-tolerant controller to deal with a severe actuator fault by leveraging a meta-reinforcement learning (meta-RL) technique. The meta-RL trains an outer-loop network to infer the faulty situation and help an inner-loop RL process to optimize the controller quickly.