Automated trajectory optimization framework for extended space missions

Long-duration space missions demand highly efficient and adaptive trajectory planning to minimize fuel consumption, extend mission life, and ensure robustness against dynamic space environments. Traditional optimization techniques often struggle with the vast and complex search spaces involved in interplanetary travel. This paper explores a hybridized approach that integrates Deep Reinforcement Learning (DRL) with Genetic Algorithms (GA) and Particle Swarm Optimization (PSO) to optimize spacecraft trajectories. DRL provides the capability to learn optimal policies in high-dimensional, continuous action spaces, while GA and PSO contribute global search capabilities and convergence efficiency. The proposed framework leverages DRL's adaptive learning with the exploratory strengths of GA and PSO to find fuel-efficient and time-optimal trajectories under multi-constraint environments. A deep neural network architecture is employed to model the policy and value functions in the DRL agent, which interacts with a high-fidelity physics-based environment simulating the space mission dynamics. GA and PSO are used to pre-train and periodically guide the DRL policy, seeding it with high-potential solutions and avoiding premature convergence. This hybrid strategy addresses the curse of dimensionality and the sparse reward challenges typical in trajectory optimization problems. This study underscores the viability and effectiveness of integrating biologically inspired optimization algorithms with modern reinforcement learning techniques for aerospace applications.