Online reinforcement learning control for aerospace systems

Reinforcement Learning (RL) methods are relatively new in the field of aerospace guidance, navigation, and control. This dissertation aims to exploit RL methods to improve the autonomy and online learning of aerospace systems with respect to the a priori unknown system and environment, dynamical uncertainties, and partial observability. In the first part of this dissertation, incremental Approximate Dynamic Programming (iADP) methods are proposed. Instead of using nonlinear function approximators to approximate the true cost-to-go, iADP methods use an (extended) incremental model to deal with the nonlinearity of unknown systems and uncertainties of the environment. In the second part, online Adaptive Critic Designs (ACDs) are proposed based on the incremental model. This method replaces the global system model approximator with an incremental model. This approach, therefore, does not need off-line training stages and may accelerate online learning. In the third part, the hybrid Hierarchical Reinforcement Learning (hHRL) method is proposed for guidance and navigation problems. This method consists of several hierarchical levels, where each level uses different methods to optimize the learning with different types of information and objectives. In conclusion, this dissertation contributes with several methods that improve the intelligence and autonomy of aerospace systems. These improvements are mainly from three perspectives: 1) enhancing the adaptability and efficiency of low-level control, 2) improving the intelligence and online learning ability of guidance, navigation, and control, and 3) creating a well-organized hierarchy to ensure coordination between each level. The proposed methods provide novel insights for both the reinforcement learning research community and for developers of aerospace automatic control system.Control & Simulatio