Genetic Algorithms Optimization of a Reinforcement Learning-based Controller for Vertical Landing Rocket Case

A reward function in reinforcement learning is the formalization of the objective. Finding the ideal reward function is a challenge, that needs a search strategy to be constructed. Genetic Algorithm is a suitable approach for reward function search due to its thoroughness. The Deep Deterministic Policy Gradient (DDPG) algorithm, which is the focus of this research, is a reinforcement learning-based controller which performances are improved after the Genetic Algorithms optimizes the agent's reward functions. The optimized controller results in narrower missed distance and lower landing velocity compared to referenced DDPG controller, and significantly less fuel consumption compared to PID.