How Artificial Intelligence Helps NASA
News Space exploration brings a set of challenges that requires a need for autonomous decision-making. Without humans, it's up to AI to fill the void.
Humans are explorers, always seeking to expand the reach of humanity and knowledge. Space is a natural frontier for humanity, telling us how Earth, our solar system and the universe formed and if we have neighbors somewhere.
The incredible distances involved in space require operations without humans in the loop.
When you can't call home
The New Horizons spacecraft flyby of Ultima Thule in January 2019 was over 4 billion miles from Earth. Its radio signals took over 6 hours to reach us. Because of limited communications stations, we had only intermittent contact. By comparison, we talk to the Mars rovers two or three times per Martian day, and Earth orbiters perhaps six times per day. In order to best use these valuable assets, spacecraft need to be able to make decisions on their own.
Artificial Intelligence (AI) has been used to support scheduling space missions for decades, going all the way back to the SPIKE system used to schedule the Hubble Space Telescope and the GPSS system used to schedule Space Shuttle preparation for reflight operations in the 1990s. Automated AI scheduling has parallels in the commercial world in AI and operations research for a wide range of fields: logistics, supply chain management, production management and pharmaceuticals, just to name a few. AI is also used to enable the Curiosity rover to select its own science targets based on scientist-selected criteria.
Sorting the stars
For any given assignment, we need to examine massive quantities of data. Machine learning techniques can triage these large data sets so that scientists can quickly examine the most interesting or novel data. AI can search these data sets to understand irregularities, common patterns and other key characteristics that lead to scientific discoveries.
In the 1990s, this SKICAT system used a machine learning technique called decision trees to classify stars and galaxies in data from the 2nd Mount Palomar Sky Survey (POSS-2). Using machine learning to deal with enormous datasets can translate directly to business problems.
Seeing Earth in a new way
AI is being used in many ways to increase our understanding of Earth’s ecosystems. AI was used to intelligently network space and ground sensors to monitor volcanoes and flooding. AI scheduling is being used to operate Earth Observing space missions such as ECOSTRESS to study plant stress and the Orbiting Carbon Observatory 3 (OCO-3) mission to measure carbon dioxide in the Earth’s atmosphere.
Machine learning is also in widespread use to solve problems such as classification of imagery to track science phenomena such as plant growth, weather, burn scar mapping, cryosphere and others. Many companies are using machine learning to develop business intelligence from this orbital imagery.
AI is a key component of the search for extraterrestrial life. To go to extreme environments such as the sub-ice oceans of Europa (a moon of Jupiter), a robotic explorer needs to travel vast distances through high radiation and ice sheets into an unexplored ocean. It must do this with little guidance from Earth. Such a robot would only be able to contact Earth every few weeks or even months. In the absence of detailed guidance from the Earth, it must navigate out and back from a base station, avoid hazards and study instrument data for minute signals of life. When a potential signature is found, it must balance the risks of further exploration against dangers to itself. Finding life is of no value if the data is not returned to Earth. Ideally, it would return the key data then return to the site for further exploration. This would be a large step for current AI technology.
An interstellar mission represents an even greater challenge for AI. Proxima Centauri and the Trappist system require a voyage of decades, fully independent from the Earth. When arriving, the probe must operate autonomously, deciding observations to take, what data to return and how to survive — a true challenge for AI.