GUIDE: Guided Updates for In-context Decision Evolution in LLM-Driven Spacecraft Operations

2026-03-28

AI Summary

Current large language models employed in spacecraft operations rely on static prompting, limiting their ability to continuously refine decision-making across tasks. This work proposes GUIDE, a novel framework that achieves the first non-parametric, in-context policy evolution by maintaining a state-conditional natural language manual of decision rules. Without updating model weights, GUIDE enables policy adaptation through iterative refinement of this rulebook, guided by an offline reflection mechanism that analyzes historical trajectories. A lightweight execution model handles real-time control, while the evolving rulebook supports long-term strategic improvement. Evaluated on adversarial orbital interception tasks in Kerbal Space Program, GUIDE substantially outperforms static baselines, demonstrating its adaptability and effectiveness in complex aerospace operations.

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 \textsc{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.

Problem

• large language models
• spacecraft operations
• in-context learning
• policy improvement
• decision evolution

Innovation

• in-context learning
• non-parametric policy improvement
• structured decision rules
• LLM-driven spacecraft operations
• cross-episode adaptation