Robust Low-Thrust Trajectory Design for Interplanetary Spaceflight: An Adaptive Latent Reinforcement Learning Method.

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Control Systems

Background:

• Spacecraft trajectory design faces challenges due to state and observation uncertainties.
• Existing methods struggle to mitigate the impact of these uncertainties on control performance.

Purpose of the Study:

• To develop a robust trajectory design method for low-thrust spacecraft under uncertainties.
• To mitigate the adverse impact of uncertainties on control performance using a novel RL approach.

Main Methods:

• An adaptive latent reinforcement learning (RL) scheme based on sequential latent variable models (SLVMs) was proposed.
• A stochastic latent proximal policy optimization (SLPPO) scheme integrated SLVM representation learning with PPO.
• An improved dense reward shaping mechanism enhanced training efficiency.

Main Results:

• The SLPPO scheme derived control policies from learned stochastic latent variables, outperforming methods using raw observations.
• Numerical simulations for two rendezvous missions demonstrated the effectiveness of the proposed approach.
• The method successfully mitigated the adverse impact of uncertainties on spacecraft control.

Conclusions:

• The proposed adaptive latent RL scheme offers a robust solution for spacecraft trajectory design under uncertainty.
• This approach enhances control performance and training efficiency in complex space missions.
• The method is validated for practical applications like rendezvous missions.