NASA and Red Hat Create AI Medical Assistant for Astronauts in Deep Space

What happened?

Researchers at NASA's Johnson Space Center, in collaboration with Red Hat, have developed a clinical decision support system called Crew Medical Officer Digital Assistant (CMO-DA). This AI-based medical assistant is designed to help astronauts diagnose and treat medical conditions during deep space missions, where real-time communication with doctors on Earth is limited or impossible.

The system relies on RamaLama, an open source tool backed by Red Hat that allows running and serving AI models predictably and securely across different hardware environments. RamaLama treats AI models as container images, using Open Container Initiative-compatible containers, ensuring portability and isolation. According to Red Hat, this approach makes AI 'boring'—that is, reliable, predictable, and unglamorous, in the best sense for mission-critical applications.

CMO-DA performs multimodal inference: it processes both large language models (LLMs) for complex medical reasoning and vision language models (VLMs) to analyze symptom images. All without needing a connection to the terrestrial cloud, which is critical for missions beyond low Earth orbit. Tests are conducted on the ground twin of HPE's Spaceborne Computer, a supercomputer that has been operating on the ISS since 2017, demonstrating the viability of high-performance computing in space.

Why is it important?

Currently, astronauts on the International Space Station (ISS) rely on doctors on Earth for real-time consultations, which is feasible due to short communication delays in low orbit (about 1-2 seconds). However, on missions to Mars, signals take between 4 and 24 minutes each way, making traditional telemedicine impossible. CMO-DA offers an autonomous solution that can save lives in critical situations, such as heart attacks, fractures, or infections, where every minute counts.

The use of open source ensures the system is reproducible, auditable, and secure—essential requirements for human safety in mission-critical environments. Additionally, the architecture could be replicated in remote areas on Earth, such as regions without internet access or limited medical infrastructure. According to NASA, more than 50% of the world's population lacks access to radiology services, and CMO-DA could help bridge that gap.

The historical context is relevant: during the Apollo missions, doctors on Earth had to guide astronauts through medical procedures with signal delays of up to 3 seconds. In the era of Artemis missions and plans for Mars, medical autonomy is an indispensable requirement. CMO-DA represents a qualitative leap from the limited medical kits and printed manuals used so far.

What consequences will it have?

Once ground tests are completed, the system will be presented to NASA leaders for potential deployment on the ISS. The next iteration will integrate Red Hat Enterprise Linux AI (RHEL AI), providing a stable and hardened foundation for scaling AI applications in extreme environments. RHEL AI includes tools like InstructLab to fine-tune models with specific medical data, improving diagnostic accuracy.

The success of CMO-DA could lay the groundwork for future autonomous medical assistants in long-duration space missions and potentially for terrestrial applications in isolated areas. It also demonstrates how open source software can be key in mission-critical applications, where reliability and transparency are paramount. Unlike proprietary systems, open source allows independent audits and customization by space agencies worldwide.

Compared to other AI-based medical assistants, such as Babylon Health's diagnostic system or IBM's Watson, CMO-DA stands out for its fully offline operation and integration with certified space hardware. Moreover, the use of OCI containers ensures the same software can run on different platforms, from HPE's Spaceborne Computer to future quantum or edge computers.

What should readers know?

• Offline operation: The system does not require an internet connection, making it ideal for remote environments. Inferences are performed locally on the Spaceborne Computer hardware.
• Underlying technology: RamaLama from Red Hat allows running AI models in containers predictably and securely. The models used include Llama 2 and versions fine-tuned for medical diagnosis.
• Hardware: Tests are conducted on the ground twin of HPE's Spaceborne Computer, a system based on AMD EPYC processors and NVIDIA GPUs, already on the ISS since 2017 and upgraded in 2021.
• Next steps: The system is expected to be tested on the ISS during the Artemis II mission (planned for 2025) and then evolve with RHEL AI to support Mars missions in the 2030s.
• Terrestrial potential: The same architecture could be used for telemedicine in areas without connectivity, such as rural regions in Africa or South America, or in disaster environments where networks are down.
• Limitations: CMO-DA does not replace a human doctor but assists the crew medical officer. Its accuracy depends on training data and is still in the validation phase.

“RamaLama makes AI 'boring'—that is, reliable, predictable, and unglamorous, in the best sense for mission-critical applications,” according to Red Hat.