GUIDE: Guided Updates for In-context Decision Evolution in LLM-Driven Spacecraft Operations
Summary: arXiv:2603.27306v3 Announce Type: replace-cross
Abstract
Large language models (LLMs) have been proposed as supervisory agents for spacecraft operations, but existing approaches rely on static prompting and do not improve across repeated executions. We introduce GUIDE, a non-parametric policy improvement framework that enables cross-episode adaptation without weight updates by evolving a structured, state-conditioned playbook of natural-language decision rules. A lightweight acting model performs real-time control, while offline reflection updates the playbook from prior trajectories. Evaluated on an adversarial orbital interception task in the Kerbal Space Program Differential Games environment, GUIDE’s evolution consistently outperforms static baselines. Results indicate that context evolution in LLM agents functions as policy search over structured decision rules in real-time closed-loop spacecraft interaction.
Introduction
The utilization of large language models (LLMs) in spacecraft operations presents significant opportunities for enhancing decision-making processes. Traditional methods have largely depended on fixed prompts, limiting the adaptability and responsiveness of these systems. The introduction of GUIDE marks a pivotal shift, allowing for dynamic improvements based on past performance.
The GUIDE Framework
GUIDE operates on the principle of non-parametric policy improvement, where the key features include:
- Cross-episode Adaptation: Unlike static models, GUIDE evolves its decision-making framework based on historical data.
- Structured Playbook: The framework utilizes a systematic approach to develop a playbook of natural-language decision rules conditioned on the current state.
- Real-time Control: A lightweight acting model is employed to ensure responsive and effective control during spacecraft operations.
- Offline Reflection: The playbook is iteratively updated through offline analysis of previous trajectories, improving decision quality continuously.
Evaluation and Results
To assess the effectiveness of GUIDE, the framework was subjected to rigorous evaluation in the Kerbal Space Program Differential Games environment, specifically focusing on an adversarial orbital interception task. The results were striking:
- GUIDE consistently outperformed static baselines in terms of decision-making efficiency and accuracy.
- The evolving nature of the playbook allowed for better contextual understanding and adaptability to changing scenarios.
- Real-time interaction capabilities enabled the spacecraft to navigate complex tasks with improved success rates.
Conclusion
The development of GUIDE represents a significant advancement in the application of LLMs to spacecraft operations. By facilitating continuous adaptation and learning from past experiences, GUIDE not only enhances the operational capabilities of spacecraft but also sets a new standard for future applications of AI in aerospace. As the field progresses, the principles established by GUIDE could pave the way for more sophisticated and autonomous decision-making systems in space exploration.