علي الله بخشي هفشجاني

This thesis surveys a novel approach to solving a problem related to orbital rendezvous of two spacecraft. In this problem, a chasing spacecraft of low – thrust type approaches a target spacecraft to be ready for docking at an appropriate distance for the mission. The target spacecraft can be a space station, a satellite in a different orbit, or two satellites with a phase difference. The proposed method tackles this problem through reinforcement learning. In this approach, a proximal policy optimization-based reinforcement learning algorithm is employed to optimize the trajectory of the chasing spacecraft. This algorithm consists of two neural networks, an actor and a critic. The actor adjusts the thrust angle based on the current state of the spacecraft and the magnitude of the thrust, which is assumed to be constant in this study, to reach a specific point in the orbit of the target spacecraft at the near-optimal time. The critic network eva‎luates the performance of the actor. In the training process, this algorithm demonstrated the capability to find a suitable trajectory for rendezvous with the target spacecraft in a simulated environment without prior knowledge of the dynamics of the environment. For the first time in this research, a policy optimization algorithm is used in reinforcement learning to solve the orbital rendezvous problem. This study investigates the changes in thrust angle and their impact on the trajectory optimization, as well as the effect of thrust magnitude on the mission completion time. Finally, the optimal values for the parameters of the reinforcement learning algorithm for training and problem-solving will be presented.

ملاقات مداري، تراست پايين ، يادگيري تقويتي، بهينه سازي سياست تقريبي،شبكه عصبي، بازيگر منتقد

orbital rendezvous, low – thrust, reinforcement learning, proximal policy optimization, neural network, actor – critic,

Ali Allahbakhshi

Majid Bakhtiari