Exploration on autopilot

The first to glimpse alien life, the first to explore another star, the first to deflect an asteroid about to hit Earth — these might sound like different heroic roles, but all might actually describe the same thing. Each milestone, if it ever does happen, could be accomplished by artificial intelligence (AI). 

We are entering a new era of space exploration, and not just because AI is set to make spacecraft more powerful. Artificial intelligence will transform the very nature of both robotic and crewed missions. Though we are only just beginning to imagine the possibilities, in the near future, more and more of humanity’s next steps in exploring the Cosmos could prove impractical, or even impossible, without AI.

Human Resources

Today, every active space mission relies on human control. Teams of experts on Earth often intervene to keep probes on track, whether planning a spacecraft's trajectory or picking a rover’s sample site. This means that the kinds of exploration demanding quick, complicated decision-making often stay out of reach. No space agency has ever flown a drone through a Martian cave, for example, though it could tell us a great deal about the planet’s past. 

This human guidance comes at a cost. Over the course of the 2010s, NASA’s Science Mission Directorate spent $2.4 billion on mission operations alone. If that work had been streamlined by just 15% 

9say, by reducing the number of hours spent managing probes) the agency could have saved $360 million — enough to pay for another mission, like the recent Double Asteroid Redirection Test (DART), in full.

Better, faster, stronger

AI changes this equation. When human control threatens to become too impractical or expensive, artificial intelligence can bridge the gap. AI will allow probes to plan out their own activities, navigate complex surroundings, and both detect and adapt to onboard problems. Missions can lean on AI for efficiency and, sometimes, to achieve entirely new feats.

Take NASA’s DART mission. In 2022, it used an onboard algorithm to guide itself to impact the asteroid Dimorphos. But unlike previous missions, DART didn't just follow a set of prewritten rules to slightly adjust its trajectory; its algorithm made its own decisions, commanding the spacecraft by itself for the mission’s entire final phase. 

Mars exploration has also benefited from machine intelligence. Starting with the Opportunity rover, every NASA rover on Mars has used a built-in system to examine its wide-field images and then automatically decide which spots to photograph in detail. The result is more photos for less work, freeing the rover teams to focus on other tasks.

But there is much more AI could do for robotic exploration. For one, experts currently plan detailed schedules for Mars rovers almost daily — but with greater onboard intelligence, future missions could instead make their own plans based on overarching goals. In one recent test at a Marslike site, a rover with this sort of “smart” software was able to complete a science campaign a full 80% faster than one using traditional methods. 

Then there's the difference AI could make in actually getting around. As it stands, spacecraft maneuvers like gravity assists and landings require extensive planning from ground control. Landing zones have to be scouted in advance, sometimes by entirely separate missions, and though rovers know to avoid some hazards, they are mostly limited to terrain that is nicely lit and easy to traverse. 

Smarter robots could one day handle all these operations on their own. In simulations, AI-driven spacecraft have planned maneuvers involving aerobraking and gravity assists that minimize fuel and travel time. AI methods have also been shown to automatically detect surface features like cave entrances, classify how navigable terrain is, and construct maps for rovers based on orbiter images. 

"In one recent test at a Marslike site, a rover with this sort of ‘smart’ software was able to complete a science campaign a full 80% faster than one using traditional methods."

A more intelligent probe might not only understand its surroundings better, but also their own workings. Research shows that AI-driven techniques can improve how a spacecraft identifies and adapts to internal problems in real time, helping missions account for broken instruments and spot software malfunctions caused by space radiation. If a wind sensor failed on NASA’s InSight Mars lander, for instance, scientists have shown they could use AI to lessen the impact on the mission's data.

I, spacecraft

Put all this together, and AI becomes perfect for self-reliant missions to unexplored worlds. 

The next generation of robotic Moon rover, ENDURANCE-R, is a prime example. By relying on onboard intelligence, ENDURANCE-R would drive up to 2,000 kilometers (1,200 miles) in just three years, covering ground at over 50 times the rate of NASA’s latest Mars rover. Though the mission has yet to be confirmed, the National Academy of Sciences has recommended ENDURANCE-R to become the agency’s top priority within one of its lunar science programs.

Planetary defense could also draw on AI to go beyond missions like DART. By using AI to pilot spacecraft, missions could hover near asteroids for long enough periods to try out untested deflection methods, like ion beams or gravity tractors. And if there is no time to send a mission to scout a dangerous asteroid beforehand, AI's ability to help a spacecraft adapt to unknown bodies could make the difference in averting catastrophe. 

Perhaps no missions stand to benefit from AI more, though, than those searching for life on faraway worlds like Europa and Enceladus. When communication signals take around an hour to travel to Earth and back, the more independent a probe is, the better. AI could also play a special role in how missions navigate unexplored terrain on these worlds, as the interview later in this issue of The Planetary Report shows.

Getting instructions to a distant lander is only half the problem, though; the other half is getting data back. Downlink speeds from faraway missions are slow, often slower than dial-up internet. In a single minute, with just one instrument, a Europa lander mission might collect 40 times more data than it could send back to Earth over its entire lifetime. Neither data compression nor transmission improvements are expected to solve this problem anytime soon.

So, upcoming missions will use AI to summarize their measurements and decide what gets sent back to scientists. The European Space Agency’s Rosalind Franklin rover will do this while looking for organic compounds on Mars, while NASA’s Dragonfly mission will similar technology as it explores Saturn’s moon Titan. Both missions are slated to launch in 2028; if either ends up finding something spectacular, AI will likely have seen it first and passed it along.

Don't say HAL

AI will make robots more capable — but it won't replace human crews. If anything, AI may be essential to expanding humanity’s presence farther from Earth. Just as more distant or complex robotic missions will depend on AI to replace mission control, space stations and crewed deep-space missions can use onboard intelligence to rely less on a faraway home planet. 

AI has already acted as a personal assistant to astronauts aboard the International Space Station, with a computer powered by IBM’s Watson AI making the trip twice in the past few years. Another assistant, this one based on a large language model similar to ChatGPT, will soon be tested by ESA on Earth.

Update in progress...

Before AI is commanding entire missions or helping out astronauts in deep space, though, there is a lot to figure out.

For one, AI methods can be hard to trust. Algorithms often function in ways that aren’t transparent, and since AI methods are meant to respond to a huge variety of conditions, testing them comprehensively is difficult. But unlike self-driving cars, an AI-enabled spacecraft can’t be crash-tested around a warehouse. Future systems will likely have to be tested almost entirely through computer simulations, and while research shows that such methods could be enough to prove that AI systems are ready to fly, this part of the field is only just getting started. 

The biggest roadblock to adopting AI for space exploration may not be issues with the technology itself, but with how it fits into space agencies. This is not new. In 1980, a panel of experts chaired by Planetary Society co-founder Carl Sagan published a report criticizing NASA’s attitude toward machine intelligence and robotics, calling it “conservative and unimaginative.” The panel argued that since computer science is so essential to space exploration, NASA should act as an incubator for it. Otherwise, the agency would have to wait for private companies to invent new technology first, then adapt it, rather than developing their own systems tailored to the needs of exploration. 

Today, NASA is no AI incubator. But it’s also not stuck in the past, and it has a very different relationship with the private sector. In 2016, well before the recent jump in AI's popularity, NASA launched an AI research partnership with Nvidia, Google Cloud, and other leading organizations. The agency has built AI methods into some recent missions, and it established its own chief AI officer in 2024. 

Still, NASA and other space agencies have done little when it comes to more broadly applying AI to spacecraft operations. Some experts argue that doing so will require a change in how agencies think about their missions. Spacecraft and instrument teams would have to work together more closely than they do now, agencies would have to invest in AI on more than a mission-by-mission basis, and officials would have to view AI as a way to improve a mission’s reliability, instead of as something that just adds risk.

As science progresses and challenges us to expand where we might boldly go, the impact AI could have on space exploration will only continue to grow. So will the reasons for agencies to make these kinds of changes. And with the right support, AI will provide a way of knowing the Cosmos unlike any we have ever had before.