Space was always supposed to be the final frontier of human exploration. It’s shaping up to be the final frontier for artificial intelligence too.
Last December, NASA’s Jet Propulsion Laboratory used Anthropic’s Claude models to help plan two Mars drives for the Perseverance rover, with human planners checking and adjusting the route before upload. In May, NASA and IBM put a compressed AI model on the International Space Station and a satellite to identify things like floods and clouds from orbit, the first model of its kind demonstrated in space. And in July, astronauts on the ISStested out a large language model to see if it could help with questions on maintenance procedures.
These experiments point to a larger shift in space engineering. For decades, engineers on Earth determined what a machine in space would do, and the machine would do exactly that. Now, researchers are testing whether non-deterministic systems like generative AI can give spacecraft more flexibility to interpret their surroundings, plan tasks, and one day make decisions for themselves. The technology is still far from trustworthy enough to hand over control of a spacecraft, but engineers are starting to ask whether they can afford not to as missions become more complex, distant, and numerous.
Why Spacecraft Need True Autonomy
Spacecraft have been operating autonomously for decades. But autonomy has never been the dominant model, in part because space engineers have prized systems whose behavior they can predict.
“Autonomy often does not have a deterministic outcome, which means that how you got into a certain situation changes the behavior,” says Robert Ambrose, the former chief of NASA’s Software, Robotics and Simulation Division. “So if you come into the same situation but from different paths, the outcome could be different. And so engineers hate that.”
Ambrose spent much of his career working on autonomous systems at NASA, including autonomy for the Orion spacecraft, NASA’s deep space and lunar orbiter spacecraft, and Robonaut 2, a humanoid robot designed to work alongside astronauts and went to space in 2011. With Orion, he saw how quickly the testing problem could multiply. Engineers had to consider not just what the spacecraft might do, but all the different ways it could have arrived at a decision.
But Ambrose says engineers found ways to manage that complexity, including automating the testing itself. “We fought the challenges of autonomy using autonomy,” he says. “That actually works.”
The need for autonomy becomes even more obvious the farther a mission travels from Earth. Ambrose points to a possible mission to Europa, Jupiter’s ice-crusted moon, where a spacecraft could dive through a water plume erupting from beneath the surface. The plume could appear too quickly for engineers on Earth to direct the spacecraft into it.
“It’s up to the spacecraft to make a decision, and we’ll be watching what happened an hour ago,” he says. A mission like that, he added, is “totally impossible” without giving the machine real autonomy.
Icarus Robotics engineers gather valuable training data for a free-flying robotic system—destined for the International Space Station—during a zero-G flight.Icarus Robotics
Adapting Robotics to Microgravity Physics
As AI and robotics advance on Earth, a commercial space boom is creating new opportunities to put those technologies to work in orbit. But what works on Earth does not necessarily work in space.
Icarus Robotics is developing what it calls a robotic labor force for space. This includes Joy, a free-flying robotic system. Joy recently finished zero-gravity testing in Canada ahead of a planned deployment to the ISS, where one of its first tasks would be moving cargo bags between modules. The company plans to start with teleoperation, using that data to eventually train the robots to work on their own.
“What the rollout will probably look like is something much closer to beginning with partial autonomy, so you still have human supervision in the loop at all times,” says Jamie Palmer, Icarus’s co-founder and CTO.
A robot trained on Earth learns from the physics of the environment around it, Palmer says, and in orbit, the physics is entirely different.
“If you take the newest Gemini robotics model, or you take the newest physical intelligence model, and you put it in zero-G there, it’s just going to fail immediately,” he says. On Earth, for example, a robot learns that when it pushes something off a table, the object falls. In orbit, it keeps moving.
That leaves Icarus with a problem the terrestrial robotics industry also faces, but in a more extreme form: There is very little real-world data from the environment where its robots will operate. The company is combining demonstrations from its robots in microgravity with simulations and tests on Earth to build its own dataset. “I wish there was” an available dataset Icarus could download and use, says Ethan Barajas, the company’s co-founder and CEO. “But there’s not today, not in a meaningful way.”
Managing Autonomous Spacecraft Risk
The challenge is not simply teaching spacecraft to act on their own. It is figuring out how to manage the risks of giving them more freedom.
“It has been mind-boggling to me how little autonomy we have in space applications,” says Ufuk Topcu, an engineering professor at The University of Texas at Austin and the director of the Center for Autonomy. “Because it’s exactly the place where human involvement is extremely hard, the stakes are high, and you need to act fast.”
Topcu says that the goal cannot be to guarantee that autonomous systems will never do anything wrong. They’re most useful in situations humans cannot anticipate, he says, so instead researchers need to start with restricted applications, learn how the systems behave, and gradually expand where and how they are used. The real question, in his view, is how well the risk of deployment is managed, not whether they can be fully eliminated.
That may become increasingly important as the space industry changes. For most of the space age, a small number of government agencies designed missions that could take decades to develop and operate. Commercial companies are now putting more spacecraft into orbit, and new missions can be developed and launched much faster.
“Space used to have very slow innovation cycles,” Topcu says. “They would think of a mission concept and spend 10 or 15 years on it. It’s not like that anymore. Everything is evolving faster now.”
- AI Seeks ET: Machine Learning Powers Hunt for Life in the Solar System ›
- NASA Let AI Drive the Perseverance Rover ›
- AI in Space ›
Facts Only
* NASA’s Jet Propulsion Laboratory used Anthropic’s Claude models to plan two Mars drives for the Perseverance rover.
* Human planners checked and adjusted the routes before upload.
* NASA and IBM placed a compressed AI model and a satellite on the International Space Station to identify floods and clouds from orbit.
* Astronauts tested a large language model on the ISS for maintenance procedure questions.
* Engineers are testing whether non-deterministic systems like generative AI can allow spacecraft more flexibility in task planning and decision-making.
* Autonomy often lacks deterministic outcomes, meaning the behavior depends on the path taken to a situation.
* Engineers managed complexity in autonomous systems by automating testing.
* A mission to Europa presents challenges where spacecraft must make real-time decisions, such as navigating water plumes before Earth intervention is possible.
* Icarus Robotics developed a robotic labor force for space, including the free-flying system Joy, for deployment to the ISS.
* Robots trained on Earth do not inherently learn from the physics of the orbital environment.
* There is a lack of meaningful real-world data for training robots in microgravity environments.
Executive Summary
Experiments involving large language models and AI were conducted in space to explore applications in space engineering. NASA’s Jet Propulsion Laboratory used Anthropic’s Claude models to assist in planning two Mars drives for the Perseverance rover, with human planners reviewing routes before upload. Further experiments included placing a compressed AI model and a satellite on the International Space Station to identify phenomena like floods and clouds from orbit, marking the first model of its kind in space. Astronauts also tested a large language model on the ISS to assess its utility for maintenance procedure inquiries.
The research points toward a potential shift in space engineering, moving from human-determined machine behavior to testing non-deterministic systems like generative AI's capacity for flexibility in interpreting surroundings and planning tasks. This transition raises questions about delegating decision-making authority to spacecraft as missions become more complex and distant.
Challenges exist regarding autonomy because the outcomes of autonomous systems can be non-deterministic, meaning different paths to a situation can yield different results, which challenges traditional engineering predictability. Furthermore, applying terrestrial robotics principles directly to space environments is complicated by the vastly different physics of microgravity, creating data scarcity for training robust autonomous systems in orbit. The overarching concern is managing the risk associated with increased autonomy rather than guaranteeing infallibility, especially given the high-stakes nature and rapid pace of commercial space development.
Full Take
The narrative traces a tension between the historical engineering preference for deterministic systems and the emerging reality of leveraging non-deterministic AI for complex, high-risk space operations. The core friction point lies in managing uncertainty: engineers resist autonomy because it introduces path dependency into outcomes, yet extreme mission complexity necessitates it. This reveals a fundamental challenge in applying terrestrial concepts to extraterrestrial environments where physical laws are profoundly different (microgravity), creating an acute data gap for training systems effectively.
The push toward autonomy is framed as necessary due to the sheer distance and complexity of future missions, suggesting that progress demands accepting risk management over perfect prediction. The argument shifts from *guaranteeing* safety to *managing* deployment risk by restricting initial application, acknowledging that AI’s utility may lie in handling unforeseen scenarios rather than flawless execution.
The dynamic described—where autonomy is sought but tempered by real-world constraints (data scarcity, physical novelty)—suggests a pattern of technological adoption where theoretical potential outpaces immediate practical capacity. The implication for human agency centers on the necessity to define acceptable risk boundaries when transferring control, especially as innovation cycles accelerate in the commercial space sector. The question becomes: how do we establish frameworks for managing and communicating the uncertainty inherent in systems that operate outside predictable physical constraints?
Bridge Questions: If safety cannot be guaranteed, what new metrics should govern the deployment of autonomous spacecraft where human oversight is delayed? How can methodologies for learning in microgravity physics be generalized to account for massive environmental shifts? What structures are necessary to manage the liability when decisions made autonomously lead to unforeseen outcomes far from Earth?
Sentinel — Human
The text reads like a high-quality synthesis of existing expert discussions regarding the application of AI and autonomy to space engineering, rather than purely generated content.