Autonomy in orbit

As humans look to expand space exploration and reach further destinations, the facilities and resources they need to survive become more complex. Missions to the Moon or Mars require environmental control and life support systems (ECLSS) for astronauts that function more autonomously given the extreme, restrictive, and unfamiliar conditions presented by deep space travel.

Currently, experts on Earth operate ECLSS on the International Space Station, analyzing data collected from systems onboard that meet the metabolic and environmental needs of the crew by providing a breathable atmosphere, potable water, and food and waste management. Because these technologies and databases are neither well integrated nor fully available onboard, the habitat is highly dependent on communications with and expertise from ground control crews.

“It’s easy to forget how much our so-called autonomous technologies rely on support and intervention by knowledgeable humans within a relatively short communication range, especially when things go wrong,” said Mario Bergés, professor of civil and environmental engineering and lead of the HOME project’s Carnegie Mellon University research team.

As one of the final installments in a five-year NASA-funded project Habitats Optimized for Missions of Exploration, or HOME, a team of researchers propose using AI-enabled digital twins to integrate dissimilar ECLSS information models and move these systems closer to autonomy. This transformation would enable the habitat to interpret its own data, answer queries, and inform decisions using mission control knowledge accessible within the smart habitat itself.

In a study published in the Journal of Aerospace Information Systems, the team introduces their vision for a digital twin framework, consisting of the physical ECLSS assets, which includes its material components and behaviors, and a supervision agent, namely a human or software onboard to perform analysis and execute control inputs. Unlike the current system on the International Space Station, where mission control on Earth handles supervision, this framework enables the agent to operate independently of ground support.

The digital twin itself is characterized by robust information and simulation models which work together to process detailed semantic information about the system and inform, predict, and modify its current and future behaviors. Using data, sensing, and AI, the digital twin is constantly informing and updating itself based on the most recent data it collected, ensuring the system is always current.

The supervision agent uses the models to complete a decision-making loop, informing the habitat and its occupants how to act in the face of anomalies or fault detection.

Our approach should be more scalable and easier to adapt to different habitat conditions and designs.

Mario Bergés, Professor, Department of Civil and Environmental Engineering

“While much of the work out there on digital twins is focused on bespoke models created from scratch, we envision a digital twin framework that ties together multiple existing digital representations of the habitat and its subsystems through a federated framework, while keeping them consistently updated via sensing and actuation loops,” said Bergés. “This approach should be more scalable and easier to adapt to different habitat conditions and designs.”

The study was conducted in collaboration with CMU’s Burcu Akinci, head of the Department of Civil and Environmental Engineering, and Ph.D. students Nicolas Gratius, Zhichen Wang, and Min Young Hwang. Other research institutions included Western New England University, University of Colorado Boulder, and University of California Davis. CMU’s contributions feed into NASA’s larger multi-university Space Technology Research Institute to advance the design of autonomous systems for space habitats.