Trajectory Optimization of Rapid Space Debris Capture Based on Deep Reinforcement Learning

With the continuous advancement of space technology, human exploration of space has become increasingly frequent. This progress has led to the development of numerous space-based products that benefit humanity and address a wide array of scientific challenges. However, these frequent space activities have also precipitated a significant deterioration of the space environment, with space debris emerging as a critical issue. The presence of rapidly moving space debris poses substantial threats to the safety of numerous spacecraft in orbit. Consequently, active space debris removal is imperative for maintaining and purifying the space environment. Trajectory optimization is a pivotal component in space debris capture missions. Throughout these missions, achieving fuel efficiency is essential to extend the operational lifespan of the capture spacecraft. Concurrently, the spacecraft must comply with stringent relative velocity and position constraints during debris capture. Traditional trajectory optimization algorithms, however, are often hampered by slow computation speeds, rendering them inadequate for meeting the real-time demands of on-orbit operations. In this paper, we investigate the trajectory optimization for the rapid capture of space debris. The capture spacecraft employs impulse thrust for maneuvering, operates with limited fuel reserves, and is constrained by impulse magnitude limits. Throughout the transfer process, the spacecraft must navigate around dense clusters of debris and precisely maneuver to the target debris. Successful capture mandates that the relative velocity and position of the spacecraft and the debris meet specific criteria. Given the dynamic nature of the space environment and its evolving conditions, it is imperative to optimize the transfer trajectory in real-time. Traditional planning algorithms are often computationally intensive and fail to meet the real-time requirements necessary for effective space debris capture. To address this challenge, we propose a trajectory optimization method based on reinforcement learning (RL). Our algorithm can make real-time decisions based on observed states after comprehensive training, functioning as an intelligent and autonomous system that obviates the need for ground support. The insights and methodologies presented in this research offer potential pathways for the intelligent and comprehensive advancement of aerospace capabilities. The main innovations of this paper are summarized as follows. First, we propose a trajectory optimization algorithm based on RL for the space debris capture problem, addressing it as a combinatorial optimization problem with multiple constraints. Our algorithm computes rapidly and is suitable for in-orbit deployment. Second, we enhance the network architecture to improve feature extraction efficiency and accelerate computation. Third, we design a novel reward function that enhances the exploration efficiency and overall performance of the algorithm. To validate the performance of our proposed algorithm, we conducted extensive simulations. To establish the superiority of our approach, we compared it against established methods, specifically the Genetic Algorithm and Particle Swarm Optimization algorithm. The experimental results substantiate the effectiveness and advantages of the algorithm proposed in this paper.