Autonomous Multi-Agent Reinforcement Learning for Resilient Navigation and Space Weather Prediction in Microgravity

Abstract

This research paper examines how Autonomous Multi-Agent Reinforcement Learning (MARL) can be used in resilient navigation and space weather prediction in space exploration. The objective is to determine how MARL performs in contrast to the conventional rule-based systems, in dynamic and anticipatable space situations. The MARL-based system is set to make spacecraft autonomously modify the paths and anticipate space phenomena, including solar flares and geomagnetic storms. The main indicators, which included latency, accuracy, robustness, and agent failure rate, were compared between the MARL-based system and a rule-based system. Simulation experiments to test how the system can cope with the changing space conditions and make real-time decisions were conducted. The MARL-based system was also able to perform better in latency (30 ms) than the rule-based system (80 ms). It also proved to be more accurate (95% in weather prediction compared to 85%) and stronger (98% compared to 75%). The MARL system had 2 % failure rate of agents; hence it was more resilient to space weather disturbances. The MARL-based system has many more benefits compared to the traditional techniques, especially in autonomous navigation and space weather forecasting. It is a promising solution to space missions with a long duration because it is real-time in making decisions, flexible, and resilient. The future of work would be based on the use of real-time information, enhancing the performance of computations and hardware testing.