How Do We Make Space Robots Autonomous?

A spacecraft closing in on another satellite begins to drift. Maybe a thruster has failed. Maybe a sensor is reporting the wrong orientation. Whatever the cause, the vehicle has another problem: it cannot wait for engineers on Earth to diagnose the malfunction before deciding what to do.

It must figure out what is wrong while continuing to fly.

That predicament captures the kind of problem Ben Riviere wants autonomous robots to solve. Riviere, an Assistant Professor at the NYU Tandon School of Engineering and the NYU Courant Institute of Mathematical Sciences, develops algorithms that allow robots to make decisions when their information is incomplete, their surroundings are uncertain and mistakes have physical consequences. Space, where machines may operate far beyond immediate human help, pushes those questions to their limits.

“You can’t put something up there that works 90 percent of the time,” Riviere says. “You really need to understand what it’s doing, why it’s doing it and know that it’s not going to fail.”

Spacecraft are already highly automated. They stabilize themselves, execute maneuvers and carry out long sequences of commands. But autonomy asks more of a machine. Instead of simply following instructions, a robot must decide how to respond when reality stops matching the plan.

“You can’t put something up there that works 90 percent of the time. You really need to understand what it’s doing, why it’s doing it and know that it’s not going to fail.”

That is particularly difficult in space because the enormous data sets powering much of modern artificial intelligence simply do not exist. Self-driving cars can learn from millions of miles of driving. Language models digest enormous collections of text. A robot operating around a satellite, comet or lunar outpost has comparatively little real-world experience to draw on.

“For space robots, we have almost no data,” Riviere says. “So the intelligence has to be much more adaptable and resilient.”

Space, however, offers an unusual advantage. Although it is hostile, its basic physics are often comparatively predictable. Newton’s laws describe an orbiting satellite far better than any model can predict the movements of pedestrians surrounding a robot on a crowded New York sidewalk.

Riviere exploits that predictability. His basic strategy is to construct mathematical models of what a spacecraft is likely to do and then use optimization algorithms to search for actions that accomplish a goal.

Two men behind a robotic arm. — *Ben Riviere (left) shows off his Microgravity Mobile Manipulato.*

*Ben Riviere (left) shows off his Microgravity Mobile Manipulato.*

“We build good predictive models and then optimize over them,” he says.

The approach sounds straightforward until something goes wrong.

In one line of research, Riviere and his collaborators have developed methods for spacecraft that can diagnose their own failures while staying out of danger. Consider a satellite approaching another object when one of its thrusters begins behaving strangely. The spacecraft may not immediately know whether the thruster is malfunctioning or whether one of its sensors is misleading it.

Rather than waiting for more data to arrive, the spacecraft can deliberately move in ways that help reveal the answer.

“What we try to do is come up with motions that help the spacecraft quickly identify whether there are faults in its system,” Riviere says. “We choose motions that provide data that can very quickly disambiguate what kind of failure happened.”

The idea turns diagnosis into a physical experiment. A spacecraft might briefly fire a thruster and compare the resulting motion with what different failure scenarios predict. Each maneuver supplies evidence.

But an experiment in orbit cannot be performed without consequences. A maneuver that provides excellent diagnostic information might also push the spacecraft toward a collision. The robot must simultaneously ask two questions: What action will tell me what is wrong? And can I perform that action safely? Riviere sees little distinction between those problems.

“Creating safe systems and creating smart systems are really the same thing,” he says. “Reasoning about what it means to be safe requires intelligence.”

Riviere’s research explores ways to make that search tractable by combining continuous optimization, which handles quantities such as thrust and position, with discrete reasoning about choices such as which maneuver or strategy to pursue.

That mixture also reflects his view of intelligence itself. A capable autonomous machine, he argues, will probably need both learned behavior and deliberate reasoning.

“There’s a little bit of memory that gives us habits and makes us smart, and there’s also reasoning on the fly and coming up with new ideas,” he says. “We’re trying to blend those two views.”His work on spacecraft fault estimation combines those tasks. The robot considers a finite set of possible failures, tests which explanations best fit its observations and searches for a trajectory that remains safe under multiple possibilities. The important step is introspection: the robot is reasoning not only about the environment but about the condition of its own sensors and actuators.

Riviere describes it as “introspecting your own health as a robot that’s really far away from any human intervention.”

The mathematics behind that decision-making often relies on tree search, a family of algorithms more commonly associated with games such as chess and Go. A computer imagines possible moves, explores their consequences and focuses its effort on the branches that appear most promising.

Robots make the problem considerably harder. A chess piece has a limited set of legal moves. A spacecraft can fire a thruster for countless durations and intensities, producing an enormous range of trajectories.

Man standing in front of powerpoint presentation.

Riviere presents his research.

Riviere’s research explores ways to make that search tractable by combining continuous optimization, which handles quantities such as thrust and position, with discrete reasoning about choices such as which maneuver or strategy to pursue.

Man standing in front of presentation — Riviere presents his research.

Riviere presents his research.

That mixture also reflects his view of intelligence itself. A capable autonomous machine, he argues, will probably need both learned behavior and deliberate reasoning.

“There’s a little bit of memory that gives us habits and makes us smart, and there’s also reasoning on the fly and coming up with new ideas,” he says. “We’re trying to blend those two views.”His work on spacecraft fault estimation combines those tasks. The robot considers a finite set of possible failures, tests which explanations best fit its observations and searches for a trajectory that remains safe under multiple possibilities. The important step is introspection: the robot is reasoning not only about the environment but about the condition of its own sensors and actuators.

Riviere describes it as “introspecting your own health as a robot that’s really far away from any human intervention.”

The mathematics behind that decision-making often relies on tree search, a family of algorithms more commonly associated with games such as chess and Go. A computer imagines possible moves, explores their consequences and focuses its effort on the branches that appear most promising.

Robots make the problem considerably harder. A chess piece has a limited set of legal moves. A spacecraft can fire a thruster for countless durations and intensities, producing an enormous range of trajectories.

The stakes extend beyond individual spacecraft. Riviere imagines robots maintaining stations, assembling structures and performing work in orbit that is expensive, dangerous or simply tedious for astronauts.

At NYU, he is collaborating with Brooklyn-based Icarus Robotics on robots intended to automate tasks in microgravity. His lab is building a “free-flying” robot that moves across an extremely flat surface on air bearings. Jets of compressed air create a cushion only a few microns thick beneath the machine, nearly eliminating friction and approximating one aspect of motion in microgravity. Mounted on top is a robotic arm.

The group is particularly interested in manipulation: how a robot can grab objects, open containers or maneuver itself while applying forces that, in space, would push its entire body in the opposite direction. One longer-term experiment is inspired by spacewalks, during which astronauts must clip themselves to handholds and move along the exterior of a spacecraft to reach a work site. A robot might eventually perform some of those movements instead.

The possibilities stretch much further. Riviere envisions machines helping construct enormous structures in orbit or participating in lunar industry, where robots could collect ice and process local resources. Microgravity manufacturing could also exploit physical processes that are impossible on Earth, where gravity causes delicate structures to sag or collapse.

“Maybe we construct giant stations in space using robotics,” he says. “Maybe we build a lunar base by collecting lunar ice and processing it. Putting humans in space is awesome, and we should keep doing it, but robotics may be a more scalable way of doing some of these things 24 hours a day.”

That vision partly explains what drew him to the field. Before graduate school, Riviere was fascinated by fusion energy and the possibility that engineering could transform technologies that sound like science fiction into ordinary infrastructure. Space robotics offered another version of the same ambition.

“I’ve always been driven by these grandiose visions of engineering,” he says. “I like the idea that if we could engineer these things, we could usher in a science- fiction future.”

Getting there will require robots that can do considerably more than execute a flawless script. A machine building a lunar outpost or repairing an orbital station will inevitably encounter situations its designers did not predict. Hardware will deteriorate. Sensors will behave strangely. Objects will move differently than expected.

The robot will have to decide what it believes, test those beliefs against the world and revise its behavior accordingly.

That may ultimately be the most important shift in space robotics. For decades, machines have extended humanity’s reach because they can go where people cannot. Riviere is working toward robots that can also make decisions where people cannot intervene quickly enough.

The farther those machines travel, the more valuable that ability becomes. Space exploration has always required sending technology into the unknown. Increasingly, the technology may have to decide for itself what to do when it gets there.

How Does Fire Change When Gravity Disappears?

Fire is one of humanity’s oldest technologies — and one of the last things anyone wants loose inside a spacecraft. On Earth, our intuition about flames is shaped by gravity: hot gases rise, cooler air rushes in below, and flames stretch upward into their familiar teardrop shape. But take gravity away, and much of that choreography disappears.

For Augustin Guibaud, Assistant Professor of Mechanical and Aerospace Engineering whose work has examined combustion in microgravity and partial gravity, that difference is more than an academic curiosity. It is a problem with consequences for astronaut safety, spacecraft design and future settlements on the Moon and Mars.

Guibaud’s research, carried out with collaborators in Europe and Japan, has focused particularly on flames spreading along insulated electrical wires. That choice is deliberate. Spacecraft are threaded with cables, and electrical faults can provide both an ignition source and combustible material. A small wire fire in a sealed vehicle, far from emergency services and surrounded by sensitive equipment, could become a serious hazard.

The first challenge is that fire in microgravity does not behave like fire on Earth. In terrestrial conditions, buoyancy continually carries hot combustion products away from a flame. In microgravity, that natural circulation nearly vanishes. Oxygen and combustion products instead move largely through diffusion and whatever airflow is produced by fans and ventilation systems.

That changes flame shape, temperature, spread rate and the production of soot and toxic gases such as carbon monoxide or hydrogen cyanide, both lethal at low concentration, which can rapidly accumulate in the constrained environment of a spacecraft.

In experiments on polyethylene-coated wires, Guibaud and colleagues have explored how forced airflow affects flames when buoyancy is absent. The work shows that even relatively gentle currents can strongly influence whether a flame grows, spreads or extinguishes. On a spacecraft, ventilation is therefore not simply a comfort system. It can become part of the combustion environment.

A group of four people standing next to an airplane on a runway. — Guillaume Legros, Axelle Bonhoure, Jean-Marie Citerne, and Augustin Guibaud pose before boarding the AirZeroG aircraft, which flies parabolic maneuvers to create brief periods of microgravity for their onboard experiments.

Guillaume Legros, Axelle Bonhoure, Jean-Marie Citerne, and Augustin Guibaud pose before boarding the AirZeroG aircraft, which flies parabolic maneuvers to create brief periods of microgravity for their onboard experiments.

One of the most intriguing threads in Guibaud’s work concerns soot. On Earth, soot is familiar as the black material deposited by incomplete combustion. These carbon-rich particles are formed inside flames, where there is little oxygen and high temperatures, and are usually destroyed as they meet and react with the oxygen of the surrounding environment. Soot particles absorb and emit thermal radiation, which is the reason why almost all fires have a characteristic yellow-to-orange glow. This glow evidences that part of the heat released by the chemical reactions in the flame escapes in all directions in the form of radiation, warming up the surrounding environment. As soot accumulates in a flame, these radiation can transfer heat toward unburned material farther along a wire, cause it to decompose, and make continued flame spread easier.

In space, the concentration of soot particles in a flame increases dramatically, so their contribution to the flame spread mechanism is expected to increase significantly.

Guibaud and his collaborators have investigated this radiative feedback in microgravity, developing methods for measuring how much energy a flame sends back toward its fuel, and how much is lost to the surrounding environment. Their results help explain why models based mainly on terrestrial convection can miss important aspects of spacecraft fire behavior.

Pressure and oxygen concentration add another layer. Future exploration vehicles or planetary habitats will not use exactly the same atmospheric pressure and composition we experience in buildings on Earth. Engineers may choose lower pressures combined with higher oxygen concentrations to reduce leakage rates, allow astronauts to breath normally, and help them put on and remove their pure oxygen spacesuit faster..

But a lower-pressure atmosphere cannot automatically be assumed to be safer from fire. Breaking down internal flame spread mechanisms, Guibaud’s studies show that pressure affects soot formation, heat transfer and flame extinction. Oxygen concentration matters as well, and the projected oxygen-rich spacecraft atmosphere suggests a dramatic increase in flammability risk. Taken together, these variables can shift the boundary between a flame that dies out and one that continues traveling along a material.

That point is especially relevant to deep-space missions. “In low Earth orbit, astronauts aboard the International Space Station remain comparatively close to Earth,” says Guibaud. Future crews traveling toward Mars could be millions of kilometers from immediate assistance. A spacecraft fire would have to be detected, understood and controlled by the crew and vehicle itself.

In a review of spacecraft fire safety, Guibaud and colleagues placed the laboratory physics in this broader context, examining past incidents and the challenges posed by deep-space exploration. The underlying problem is stark: fire-safety standards developed under Earth gravity do not necessarily reproduce the environment in which a spacecraft fire would actually occur.

The same concern extends to smoke detectors. More recent research involving Guibaud has investigated how smoke particles develop and travel in low gravity. Their size, concentration and optical properties determine whether spacecraft sensors can detect a fire quickly. If microgravity changes the particles themselves, then detectors optimized using terrestrial smoke may not always respond in the same way in space.

And microgravity is only part of the story. The Moon has about one sixth of Earth’s surface gravity; Mars has roughly three eighths. Experiments in partial gravity suggest that flames do not necessarily change smoothly between zero gravity and Earth gravity. Researchers have observed cyclic flame-spread behavior and even extinction regimes under intermediate gravitational conditions they did not anticipate.

That means a Martian fire cannot simply be predicted by taking an Earth result and scaling it downward.

Recent work involving Guibaud has pushed this idea further by examining the interaction between pressure and gravity-induced “residence time”, which is the length of time gases and soot-forming material remain in hot regions of a flame. As gravity weakens, combustion products can linger longer. At the same time, raising pressure can increase soot production. Both mechanisms can push a flame toward a visible smoke-producing state, but for different physical reasons.

The emerging picture is less a single rule about “fire in space” than a network of interacting effects which need to be carefully balanced to achieve safety. Gravity alters flow. Flow controls oxygen delivery and residence time. Pressure influences reaction rates and soot. Soot modifies radiation. Radiation heats fuel. And the changing atmosphere around a flame can feed back on the entire process. For future spacecraft designers, that complexity is inconvenient, but necessary to understand.

Guibaud’s work illustrates a larger lesson of human space exploration: technologies that seem thoroughly understood on Earth can become unfamiliar once gravity changes. A wire still looks like a wire. Oxygen is still oxygen. A flame is still a chemical reaction between fuel and oxidizer. Yet the physics connecting them can shift enough to create new risks.

As astronauts prepare to spend longer periods beyond Earth, the safest spacecraft may be those designed not around our terrestrial intuition about fire, but around what flames actually do when “up” and “down” begin to disappear.

How Do We Help Astronauts Get a Good Night’s Rest?

Astronauts aboard the International Space Station can see the sun rise and set roughly 16 times a day. It is a spectacular view, but for the human body it presents a problem. Our internal clocks evolved under a dependable 24-hour cycle of daylight and darkness. In orbit, that environmental signal becomes profoundly scrambled.

For biomedical engineer Devjoy Dev, the problem raises a deceptively simple question: What happens to sleep when gravity, one of the body’s most constant physical cues, nearly disappears?

Dev explored that question during his doctoral research at NYU Tandon School of Engineering, where he worked with Khalil Ramadi on the connections among microgravity, circadian rhythms, the gut and the brain. The work began as an investigation into astronaut sleep, but it gradually developed into a broader effort to understand whether electrical stimulation of the gastrointestinal system might eventually provide a new way to influence sleep and other biological processes.

Dev did not originally plan to study either spaceflight or the gut. Trained as a biomedical engineer, he studied medical devices, implant design and tissue engineering at Queen Mary University of London before moving into cardiovascular technology. His early work spanned cardiac device design, imaging and technology transfer. At NYU, he brought that background in electrophysiology and electrical stimulation to an entirely different organ system.

The gut, Dev quickly learned, is far more than plumbing.

“The gut is this weird amalgamation of tubes,” he said, but it is also “so advanced, it’s so intelligent, and it’s doing way more than just digestion.”

That complexity is partly why the gastrointestinal tract has become an increasingly attractive target for bioelectronic medicine. Neural and chemical signals continually travel between the digestive system and the brain, helping regulate metabolism, mood, immune function and other processes. Dev became interested in whether that communication network could also offer a way to influence sleep.

Spaceflight supplied an extreme environment in which to investigate the problem.

Astronauts often sleep less than they would on Earth. One reason is the disruption of the external cues that help synchronize circadian rhythms, especially sunlight. Temperature, meal timing and other signals can serve as secondary cues, but space adds still more complications, including isolation, radiation exposure and microgravity. In a setting where crews operate complex equipment and make consequential decisions, chronic fatigue can become more than an inconvenience.

“If you’re cognitively compromised,” Dev said, “then that’s posing risks not just on the mission, but also on your lives and your crewmates’ lives.”

Dev focused particularly on microgravity. To study its effects without leaving Earth, he used a random positioning machine, a device that continuously rotates biological samples around multiple axes. Rather than actually eliminating gravity, the machine changes the samples’ orientation so rapidly that the cells experience an environment designed to approximate some biological effects of microgravity.

Two astronauts in the International Space Station sleeping vertically.

Commander Richard “Dick” Truly and Mission Specialist Guion Bluford sleep on Challenger’s middeck. Photo courtesy of NASA.

Two astronauts hold up a bag of materials.

Astronauts Kathryn D. Sullivan and Sally K. Ride display a “bag of worms,” a sleep restraint. Photo courtesy of NASA.

He then examined cells whose circadian rhythms had been experimentally synchronized and tracked several well-known clock genes, including PER1, PER2 and BMAL1. On Earth, these genes typically rise and fall in characteristic patterns over the course of a day. Those molecular oscillations are part of the biological machinery that helps coordinate sleeping, waking and countless other physiological functions.

Under simulated microgravity, some of those patterns changed.

The shifts were not necessarily immediate. When Dev sampled cells at intervals across a single day, the circadian system appeared relatively resilient. But when he compared the cells over periods of 24, 48 and 72 hours, larger differences emerged. Some clock genes remained stable at first and then changed after prolonged exposure.

“The longer the microgravity progresses, the more impact it’s gonna have, or change it’s gonna have, on our circadian gene expression,” Dev said.

The findings were an early step, not a complete model of astronaut sleep. The experiments were conducted in cell and tissue culture, and Dev says future work would need to examine a wider range of circadian biomarkers, extend experiments beyond three days and move into more realistic three dimensional systems such as organoids.