ÎÞÓÇ´«Ã½

Deploying a Large Language Model in Space

The Challenge: Erase Compute Limitations Beyond Earth

The ISS orbits Earth about once every 90 minutes at an altitude of approximately 250 miles while reaching speeds of up to 17,500 miles per hour, or 5 miles per second. These operating conditions place severe demands on the high-bandwidth geosynchronous network architecture required to support the station’s multifaceted investigations.

In the U.S. module alone, 100 data networks transfer hundreds of thousands of signals. But as ISS personnel contend with constraints on network coverage, bandwidth, and latency, their essential work can be hindered by link delays and disruptions. By automating and augmenting data collection and analysis, the ISS would recover crew time and improve overall crew capability. It also sought to test out new technologies needed to explore beyond low Earth orbit where storage, processing, and communications become less available and more constrained.

To help our clients better harness data insights on orbit, ÎÞÓÇ´«Ã½ is developing on-board AI processing solutions with the potential to reliably deliver robust compute and analytics resources across the challenging edge environment of space.

The Solution: Deploy Powerful Edge AI on Spacecraft

promise to revolutionize the data-intensive space domain. Generative agents instantly augment human analysis, search and summarization, and communication capabilities, with the potential for enhancing the speed of command-and-control decisions and the precision of satellite maneuvers. Transitioning LLMs from conventional cloud-based environments to the edge where users collect and often store data can dramatically improve latency and other technical requirements.

To move closer to realizing these benefits, the ÎÞÓÇ´«Ã½ team re-architected and deployed the first lightweight LLM application for space. Optimized to operate in space-based environments such as the ISS, ÎÞÓÇ´«Ã½â€™s LLM incorporates RAG, a technique that enhances the relevance and accuracy of LLM responses by using a database search to enrich prompt results.

The application provides an inverted index of relevant texts to improve processing efficiency and allow for longer queries. The application design also speeds up queries and eliminates the need for a power-hungry vector database. And templates simplify prompt engineering, further refining the user experience.

Bolstered by these state-of-the-art model techniques, our on-board LLM solution will enable personnel to search for and efficiently retrieve relevant information from a corpus of instruction manuals and operational documents using natural language queries. ÎÞÓÇ´«Ã½ technologists optimized and containerized the model to run reliably in the resource-constrained ISS compute environment, where only solar panels provide energy, overheating computers can create hazards, and devices must be compact. Working with HPE, our team deployed the LLM by up-linking it to the HPE Spaceborne Computer 2 (SBC-2) on board the ISS. Working in partnership with the ISS National Lab, HPE has positioned the most powerful computer to ever fly to space to support this type of advanced research and development (R&D).

Putting the LLM onto orbit represents a significant ÎÞÓÇ´«Ã½ investment in technology enablement as we continue building our R&D portfolio for space. The portfolio extends our more than 60-year record as a space program partner for defense, intelligence, and civil agencies to catalyze innovation in space domain awareness, advanced ground systems, system-of-systems engineering, and more.

The Impact: Harness Data Insights Where Space Missions Lead

With ÎÞÓÇ´«Ã½â€™s support, the ISS and future manned space systems are positioned to benefit from LLM capabilities that augment human expertise and directly support research and exploration activities. While there is never a complete replacement for an astronaut-in-the-loop, we believe instantaneous human augmentation with an LLM RAG can help astronauts, and other terrestrial–based experts in constrained, stressed environments, achieve their missions.

Our on-board LLM is designed to streamline worker access to vetted technical knowledge and create opportunities for faster, more accurate decisions. What’s more, the LLM provides the basis for future innovations, such as a personalized, virtual, subject-matter expertise that users can carry in their pockets; multimodal AI tools; and meshed LLMs that provide automated decision support in complex scenarios.

U.S. intelligence community and DOD organizations are also interested in using AI to perform inference and analysis on board satellite buses in ways that augment ground-based analysis. Driving this mission requirement are the compute, latency, and bandwidth limitations that also exist on the ISS.

Responding to this mission challenge, ÎÞÓÇ´«Ã½ is strategically investing to continue developing, training, and deploying AI/machine learning algorithms and LLMs tailored for specific applications in satellite operations. In line with evolving space technology needs, these applications include intelligence, surveillance, and reconnaissance (ISR); software-defined architecture (SDA); mission management/command and control (MM/C2); and autonomous maneuvers.

When milliseconds matter and resilience is a deterrent, on-board edge AI capabilities like these can reduce two-way latency from minutes to nanoseconds and increase bandwidth speed 1,000-fold, potentially making the difference for mission success.

What’s Next

Future on-board AI processing solutions from ÎÞÓÇ´«Ã½ promise to provide flexible capabilities in space at a fraction of the cost of current legacy systems; enable assets that can maneuver without being seen by adversaries; provide cyber data pipelines and custom threat alerts for space-hardened infrastructure; and scale wearable, human performance tools to manage crew readiness—all with robust technical performance and mission resilience at the austere edge of space. We will continue to take what we learn from the continued deployment of our AI capabilities in space and apply that innovation to help our commercial, civil, and DOD and intelligence community clients thrive in ground-based, maritime, and other extreme edge environments, as well as space.

Ìý

Learn more about how ÎÞÓÇ´«Ã½ can help your organization apply emerging technology to overcome space domain challenges.