Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Language Models are Spacecraft Operators

Victor Rodriguez-Fernandez, Alejandro Carrasco, Jason Cheng, Eli Scharf, Peng Mun Siew, Richard Linares

TL;DR

This paper investigates using Large Language Models as autonomous spacecraft operators for the Kerbal Space Program Differential Games (KSPDG) when traditional reinforcement learning is data-limited. By combining prompt engineering, observation augmentation, few-shot prompting, and fine-tuning with human gameplay data, the authors built a pure LLM-based agent that achieved 2nd place in KSPDG and demonstrated the feasibility of language-driven decision-making for real-time control. They show that augmenting prompts with derived telemetry and using Chain-of-Thought prompting improves generalization and reduces failures, while fine-tuning further reduces latency and improves reliability. The work highlights a practical pathway for applying LLMs to space autonomy and outlines future directions, including evaluating other models, incorporating multimodal inputs, code-generation for agents, and considerations for on-board deployment in future work.

Abstract

Recent trends are emerging in the use of Large Language Models (LLMs) as autonomous agents that take actions based on the content of the user text prompts. We intend to apply these concepts to the field of Guidance, Navigation, and Control in space, enabling LLMs to have a significant role in the decision-making process for autonomous satellite operations. As a first step towards this goal, we have developed a pure LLM-based solution for the Kerbal Space Program Differential Games (KSPDG) challenge, a public software design competition where participants create autonomous agents for maneuvering satellites involved in non-cooperative space operations, running on the KSP game engine. Our approach leverages prompt engineering, few-shot prompting, and fine-tuning techniques to create an effective LLM-based agent that ranked 2nd in the competition. To the best of our knowledge, this work pioneers the integration of LLM agents into space research. Code is available at https://github.com/ARCLab-MIT/kspdg.

Language Models are Spacecraft Operators

TL;DR

This paper investigates using Large Language Models as autonomous spacecraft operators for the Kerbal Space Program Differential Games (KSPDG) when traditional reinforcement learning is data-limited. By combining prompt engineering, observation augmentation, few-shot prompting, and fine-tuning with human gameplay data, the authors built a pure LLM-based agent that achieved 2nd place in KSPDG and demonstrated the feasibility of language-driven decision-making for real-time control. They show that augmenting prompts with derived telemetry and using Chain-of-Thought prompting improves generalization and reduces failures, while fine-tuning further reduces latency and improves reliability. The work highlights a practical pathway for applying LLMs to space autonomy and outlines future directions, including evaluating other models, incorporating multimodal inputs, code-generation for agents, and considerations for on-board deployment in future work.

Abstract

Recent trends are emerging in the use of Large Language Models (LLMs) as autonomous agents that take actions based on the content of the user text prompts. We intend to apply these concepts to the field of Guidance, Navigation, and Control in space, enabling LLMs to have a significant role in the decision-making process for autonomous satellite operations. As a first step towards this goal, we have developed a pure LLM-based solution for the Kerbal Space Program Differential Games (KSPDG) challenge, a public software design competition where participants create autonomous agents for maneuvering satellites involved in non-cooperative space operations, running on the KSP game engine. Our approach leverages prompt engineering, few-shot prompting, and fine-tuning techniques to create an effective LLM-based agent that ranked 2nd in the competition. To the best of our knowledge, this work pioneers the integration of LLM agents into space research. Code is available at https://github.com/ARCLab-MIT/kspdg.
Paper Structure (10 sections, 8 figures, 4 tables)

This paper contains 10 sections, 8 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Backgrounds
  3. KSPDG: Kerbal Space Program Differential Games
  4. Large Language Models
  5. OpenAI's ChatGPT API
  6. Employed strategies
  7. Prompt engineering and observation augmentation
  8. Few-shot prompting
  9. Fine-tuning
  10. Conclusions and future work

Figures (8)

  • Figure 1: Overview of the proposed approach to use a LLM (e.g. ChatGPT) as an autonomous spacecraft operator that gets, as user prompt, the current status of the mission from the KSDPG simulation environment (i.e., the state or observation in the RL jargon), and replies with a reasoned action to carry out, expressed as a function calling with the specific throttle vector and the textual justification behind the action.
  • Figure 2: Kerbal Space Program Differential Games. In this work we will focus on the Pursuer-Evader scenario.
  • Figure 3: Example user prompt to the LLM, showing prompt augmentation
  • Figure 4: Example manual response for few-shot prompting
  • Figure 5: System prompt used for eliciting Chain of Thought in this model
  • ...and 3 more figures