Autonomous Thrust Vector Control Using Machine Learning in Physics Simulations: Enhancing Performance in Guidance

Autonomous thrust vector control systems are a computational challenge in the field of guidance systems. It involves algorithms that are capable of autonomous management of thrust vector and thus, the trajectory of a rigid body to achieve the desired results while minimizing fuel consumption. The primary objective of the paper is to address this problem through the application of machine learning. By analyzing sensor readings, the simulated model will autonomously determine the appropriate thrust vector and thrust force, effectively establishing a closed-loop control system. Models based on quality learning are compared in a randomized simulated environment. The agents are scored based on the final position, angle, velocity, and time taken (meant to represent fuel consumption) of the body after a given time. The results indicate that in a simulated environment, reinforcement learning is appropriate for training an agent capable of autonomous thrust vector guidance. This agent can be utilized within a closed-loop control system. Additional internal or external parameters can be introduced to further enhance its capabilities.