Developing Artificially Intelligent Agents to Support Earth Independent Medical Capabilities During Human Exploration-Class Space Missions

Authors: William R. Buras, David C. Hilmers

Abstract

Artificially-intelligent agents are being developed to support NASA crewmembers during exploration missions where Earth-based medical support is unfeasible. This article describes the need for such agents, provides a functional overview, and presents a concept of operations demonstrating how they could support crew health. Desirable characteristics to support crew medical officers are listed. This technology also has potential applications in remote terrestrial environments where expert consultation is inaccessible.

Introduction

Exploration-class missions to the Moon, Mars, asteroids, and beyond present unique medical challenges not encountered during low Earth orbit (LEO) missions. On the International Space Station (ISS), astronauts maintain constant, real-time contact with Mission Control Center (MCC), allowing immediate consultation with specialists for diagnosis and treatment. Medical evacuation and resupply are possible within hours or days. However, during deep space missions, many of these capabilities vanish. Communication delays to Mars can reach 45 minutes round-trip, eliminating real-time support. Resupply and evacuation become nearly impossible. Mass and volume constraints limit onboard medical equipment. Crews must become increasingly self-reliant, managing medical problems with limited resources — a paradigm known as Earth Independent Medical Operations (EIMO) [1]. To enable EIMO, crews must substitute onboard expertise for MCC support. However, crew medical officers (CMOs) have limited preflight training, and skills degrade during long missions. Cognitive challenges such as “space fog” further complicate medical decision-making [2,3]. Advances in artificial intelligence (AI) offer promising tools to augment crew capacity in training, diagnosis, treatment, and inventory management. These include mixed reality for just-in-time training (JITT), AI-driven chatbots using large language models (LLMs), and onboard reference materials. Imagine a CMO on a Mars mission faced with a crewmember experiencing chest pain. With a 45-minute communication delay to Earth, how can AI assist in diagnosis and management using limited onboard resources? Key qualities for such an AI system include [4–6]:

• Interactive questioning and recommendations like a ground-based consultant.
• Fully autonomous operation without real-time external assistance.
• Compliance with strict mass, volume, power, and data constraints.
• Comprehensive medical database covering common and rare conditions.
• Symptom-based dialog guiding history-taking, examination, and testing.
• Optimized, logically progressive questioning.
• Recognition of emergent conditions requiring immediate treatment.
• Procedural guidance and just-in-time refresher training.
• Communication tailored to operator expertise level.
• Continuous background monitoring and alerting.
• Instant responses to medical queries.
• Hands-free voice interaction.
• Detailed knowledge of crew medical history and onboard resources.

Methods

Agents in Development

Our team is developing AI tools to meet EIMO needs, enabling astronauts to independently manage medical problems during exploration missions. Central to this effort is Space Medicine GPT (SGPT), a localized, distilled LLM designed to fulfill needs #1–7 and #9–13. Unlike general chatbots, SGPT uses novel prompt engineering to create an interactive question-answer environment to reach diagnoses. SGPT operates offline with a targeted size under 50 billion parameters, suitable for laptops, smartphones, and tablets, addressing mission constraints. It integrates vector databases for rapid retrieval of evidence-based medical resources. Training includes physician-patient conversation data to emulate expert consultants. SGPT guides the CMO through symptom history, physical exams, and tests, applying Bayesian logic [8] to prioritize questions and rule out emergencies like acute coronary syndrome or tension pneumothorax (see Fig. 1). Fig. 1: Conceptual Framework for AI-assisted onboard medical diagnosis and treatment during Earth Independent Medical Operations. SGPT adapts communication to the CMO’s expertise, continuously monitors vital signs via wearables, and provides instant alerts. It also serves as a conventional medical information source. To support procedural guidance and refresher training (#8), we developed the Intelligent Medical Crew Agent (IMCA), which uses augmented reality (AR) for step-by-step medical procedures. AR devices project imaging, diagrams, checklists, and videos tailored to operator skill levels. Procedures like pericardiocentesis and chest tube placement are scripted and can be rapidly updated. A key IMCA component is the Visual Ultrasound Learning, Control and Analysis Network (VULCAN). Ultrasound is the primary imaging modality onboard. VULCAN monitors operator execution in real-time, providing corrective feedback using AI, computer vision, and telemetry to ensure proper performance.

Challenges and Aspirations

While AI tools like SGPT, IMCA, and VULCAN promise to empower CMOs on missions to Mars, significant challenges remain. The complexity of clinical decision-making is difficult to emulate, requiring training on best medical practices to avoid errors. Offline operation demands distillation of large medical databases into manageable sizes. Hardware and software maturity must improve for practical deployment. Rigorous testing in Earth analogs such as Antarctica, wilderness medicine, and military deployments is essential. AI tools are only part of a comprehensive clinical decision support system, which must also include pharmacy, supply management, health maintenance, and environmental control. Despite challenges, we anticipate AI-based expertise residing on tablets or cellphones will revolutionize space medicine. Moreover, these technologies could transform healthcare in remote terrestrial settings lacking internet or expert access, such as disaster zones or war areas, saving lives and reducing suffering both in space and on Earth.

Data Availability

All data presented are available within the figures of this manuscript.

Code Availability

The code used is proprietary and confidential. Interested researchers may contact william.buras@tietronix.com for collaboration under a non-disclosure agreement.

References

• Levin, D. R., et al. Enabling Human Space Exploration Missions Through Progressively Earth Independent Medical Operations (EIMO). IEEE Open J. Eng. Med. Biol. 4, 162–167 (2023).
• Manzey, D. & Lorenz, B. Mental performance during short-term and long-term spaceflight. Brain Res. Rev. 28, 215–221 (1998).
• Welch, R., Hoover, M. & Southward, E. F. Cognitive performance during prismatic displacement as a partial analogue of ‘space fog’. Aviat. Space Environ. Med. 80, 771–780 (2009).
• Waisberg, E. et al. Challenges of artificial intelligence in space medicine. Space: Sci. Technol. (2022).
• Russell, B. K. et al. The value of a spaceflight clinical decision support system for earth-independent medical operations. npj Microgravity 9, 46 (2023).
• Garcia-Gomez, J. M. Basic principles and concept design of a real-time clinical decision support system for managing medical emergencies on missions to Mars. arXiv:2010.07029v2 (2021).
• Taipalus, T. Vector database management systems: fundamental concepts, use-cases, and current challenges. Cogn. Syst. Res. 85, 101216 (2024).
• Gill, C. J., Sabin, L. & Schmid, C. H. Why clinicians are natural Bayesians. BMJ 330, 1080–1083 (2005).
• Amador, A. R. et al. Enabling space exploration medical system development using a tool ecosystem. 2020 IEEE Aerospace Conference (2020).

Acknowledgements

We thank NASA and TRISH for research support under NASA SBIR grants NNX16CC522P, NNX17CC12C, 80NSSC120C0541, 80NSSC21C0578, and 80NSSC23PB612.

Author Information

• William R. Buras, Tietronix Software Inc., Houston, TX, USA
• David C. Hilmers, Baylor College of Medicine, Translational Research Institute for Space Health (TRISH), Houston, TX, USA

Contributions

W.R.B. led conceptualization, project administration, funding acquisition, and drafting. D.C.H. contributed to conceptualization, figure construction, and manuscript revisions. All authors approved the final manuscript.

Ethics Declarations

The authors declare no competing interests.

---

Source: Originally published at Nature npj Microgravity on 02 August 2025.

Frequently Asked Questions (FAQ)

AI Agents for Medical Support in Space Missions

Q: What is the primary challenge addressed by developing AI agents for space missions? A: The primary challenge is providing effective medical support to astronauts during long-duration space missions where real-time communication with Earth-based medical professionals is impossible due to communication delays and the unfeasibility of immediate evacuation or resupply. Q: What is Earth Independent Medical Operations (EIMO)? A: EIMO refers to the paradigm where crews must become increasingly self-reliant in managing medical problems with limited onboard resources, substituting onboard expertise for Mission Control Center support. Q: What are the key capabilities of the Space Medicine GPT (SGPT) agent? A: SGPT is designed to act as an interactive, offline medical consultant, providing diagnostic guidance, treatment recommendations, and inventory management. It can guide through symptom history, physical exams, and tests, adapting communication to the operator's expertise level and providing instant alerts based on vital signs. Q: How does the Intelligent Medical Crew Agent (IMCA) assist astronauts? A: IMCA utilizes augmented reality (AR) to provide step-by-step guidance for medical procedures, including visual aids and real-time feedback, to support refresher training and procedural execution. Q: What is the role of VULCAN within the IMCA system? A: VULCAN (Visual Ultrasound Learning, Control and Analysis Network) specifically monitors and provides corrective feedback on ultrasound procedures, utilizing AI, computer vision, and telemetry to ensure proper operator performance. Q: What are some of the main challenges in developing these AI medical agents? A: Key challenges include accurately emulating complex clinical decision-making, distilling large medical databases into manageable sizes for offline operation, and ensuring the maturity of hardware and software for practical deployment. Rigorous testing in Earth analogs is also crucial. Q: Beyond space exploration, where else could these AI medical technologies be applied? A: These technologies have potential applications in remote terrestrial environments where expert medical consultation is inaccessible, such as disaster zones or war areas.

Crypto Market AI's Take

The development of AI-driven medical support systems for space exploration, as detailed in this article, highlights a broader trend of AI integration into specialized and critical fields. Our platform, Crypto Market AI, leverages AI and machine learning for similar purposes within the financial sector, providing sophisticated tools for market analysis and trading. The concept of creating localized, distilled AI models like SGPT, designed to operate efficiently within strict constraints, is directly analogous to the development of our own AI Agents and trading bots, which are optimized for performance and deliver actionable insights without requiring constant external oversight. The emphasis on robust databases, logical reasoning (like Bayesian logic mentioned for SGPT), and adaptive communication mirrors the core functionalities of our AI analysts and trading bots, aiming to empower users with intelligent decision-making tools.

More to Read:

• AI Agents: Capabilities, Risks, and Growing Role
• AI-Driven Crypto Trading Tools Reshape Market Strategies
• Understanding Blockchain: A Comprehensive Beginner-Friendly Guide