 |
International Society of Science and Applied Technologies |
| A Deep Reinforcement Learning Approach to Dynamic Extravehicular Maintenance Policy in Space Stations | ||||
| Author | Haochen Luo
|
|||
| Co-Author(s) | Qin Zhang; Yu Liu
|
|||
| Abstract | The long-term stable operation of space stations is critical for advancing space research and technology validation, making efficient extravehicular maintenance a pivotal concern. This article proposes a multi-agent collaborative model for extravehicular maintenance, optimized using deep reinforcement learning (DRL), handle risk constraints, and adapt to dynamic environments in space environment. Utilizing Deep Q-Networks (DQN), the model integrates hierarchical maintenance planning and reactive execution agents, coupled with dynamic state transition models for device degradation, environmental hazards, and personnel stamina decay. This framework optimizes maintenance policy for cost efficiency and safety under constraints of limited spare parts, restricted extravehicular maintenance duration, and environmental risks. These insights highlight the model’s potential for cost reduction and rapid response to stochastic failures of components, ensuring astronaut safety and providing decision support for extravehicular maintenance.
|
|||
| Keywords | Extravehicular Maintenance, multi-agent system, deep reinforcement learning, dynamic maintenance policy | |||
| Article #: RQD2025-354 | ||||
Proceedings of 30th ISSAT International Conference on Reliability & Quality in Design |