Spacecraft That Think: Inside the 2025 Early Career Faculty Awards

Here’s a hard truth about space exploration: we’ve been flying blind. Not literally, of course — spacecraft are packed with sensors and cameras. But when you’re hurtling toward Mars at 12,000 miles per hour, waiting for a signal from Earth that takes 14 minutes to arrive, you’re not really driving. You’re firing a bullet and hoping it hits the target. The 2025 Early Career Faculty (ECF) Awards from the agency’s research directorate are quietly rewriting that script — and they’ve just handed the keys to three researchers who plan to make spacecraft think for themselves.

Announced in March 2025, the ECF Awards recognize early-career scientists and engineers tackling the agency’s most stubborn technical problems. This year’s cohort is small but ferocious: three projects spanning high-enthalpy diagnostics, machine learning for onboard guidance, and autonomous spacecraft planning. These aren’t incremental tweaks. They’re fundamental rethinks of how we approach atmospheric entry, navigation, and decision-making in deep space.

Burning Up at 7,000 Degrees — And Liking It

When a spacecraft slams into Earth’s atmosphere at orbital velocity, the air in front of it compresses so violently it turns into plasma. Temperatures hit 7,000 degrees Celsius — hotter than the surface of the Sun. We call this “high-enthalpy” flow, and it’s a nightmare to measure. Current ground-based test facilities use arc jets or shock tunnels, but they struggle to replicate the precise conditions of hypersonic entry. Sensors degrade. Data gets fuzzy. And critical design decisions get made on incomplete information.

Dr. Aris Beth Simoneau, a newly minted assistant professor at the University of Colorado Boulder, thinks we can do better. Her ECF project, “Advanced Diagnostics for High-Enthalpy Test Facilities Simulating Spacecraft Atmospheric Entry,” aims to build diagnostic tools that can withstand these infernal conditions — and actually deliver clean data. She’s combining laser-induced fluorescence with high-speed spectroscopy to track chemical species in real time as they form and decay.

“Right now, we’re essentially guessing at the wall heat flux,” says Dr. Simoneau. “If we can get direct measurements of the flow chemistry, we can validate our models and improve heat shield design. It’s the difference between flying a capsule with a tinfoil shield and one that’s been engineered for the real environment.”

The technology also has implications for sample return missions — like the one that recently delivered rocks from an asteroid, or future missions to bring back materials from Mars. (For a stunning example of how we caught a fleeting cosmic visitor, check out The Cosmic Drift.) If we can’t test entry conditions accurately on Earth, we risk sending billion-dollar payloads into a plasma bath they weren’t designed to survive.

Teaching Spacecraft to Navigate — Without a GPS

Here’s a question that keeps mission planners up at night: How do you land a rover on Mars when you don’t know exactly where you are? GPS doesn’t work beyond Earth orbit. Instead, spacecraft rely on star trackers, inertial measurement units, and ground-based radiometric tracking. But that chain of trust has a weak link — latency. By the time a signal from Earth arrives, the spacecraft has already moved. Every correction is reactive, not predictive.

Enter machine learning. Dr. Satoshi Yamada, assistant professor at the University of Texas at Austin, is using his ECF grant to develop “Machine Learning Methods to Enable Onboard Guidance, Navigation, and Control.” His approach? Train neural networks to recognize terrain features from orbital imagery, then use those features to localize the spacecraft in real time. Think of it as giving a Mars rover a photographic memory — it snaps a picture of the crater below, compares it to a database of Martian surface maps, and computes its position within meters.

But there’s a catch. Neural networks are notorious for being black boxes. You feed them input, they spit out an answer, and nobody really knows why. In aerospace, that’s unacceptable. So Dr. Yamada is building “interpretable” machine learning models — ones that can explain their reasoning to human operators. “We’re not replacing human judgment,” he says. “We’re augmenting it. The spacecraft says, ‘I think I’m here, and here’s why.’ The operator can then verify or override.”

This matters for more than just Mars. Future missions to Europa, Titan, or even interstellar probes will face communication delays measured in hours or days. They’ll have to make decisions autonomously. The ECF program is laying the groundwork for that capability now — before we find ourselves screaming at a spacecraft that can’t hear us.

And it’s not just about deep space. Autonomous navigation has direct applications here on Earth — think of drones navigating GPS-denied environments, or self-driving cars in tunnels. Dr. Yamada’s work could spill over into commercial technology faster than anyone expects. (As a reminder that planetary dynamics can be just as unpredictable, consider Venezuela’s shifting ground as revealed by NISAR radar data.)

Planning for the Unexpected — On a Planetary Scale

The third ECF project tackles a problem every road trip veteran knows: stuff goes wrong. On Earth, you pull over and check the map. In space, the map is obsolete the moment you launch, and pulling over isn’t an option. Dr. Elena Voss, assistant professor at the University of Washington, is building a system called “Autonomous Planning for Spacecraft Using Machine Learning Methods.” Its job? Generate and execute mission plans in real time, adapting to failures, unexpected terrain, or shifting science priorities.

Currently, spacecraft operate on pre-loaded sequences. Engineers on Earth plan every command days in advance. If a rover’s wheel gets stuck in sand — as happened with Spirit in 2009 — the team on the ground races to upload a new sequence before the rover’s batteries drain. Sometimes they succeed. Sometimes they don’t.

Dr. Voss’s system flips that model. The spacecraft itself becomes the planner. It assesses its state, scopes out the environment, and decides what to do next. “Think of it like a chess player who can’t phone their coach,” she explains. “They have to evaluate the board, see three moves ahead, and commit. The spacecraft will do the same — but with a million lines of code and a radiation-hardened processor.”

Early tests in simulated Martian terrain show the system can reduce planning time from days to minutes. That’s the difference between a rover that dies in a dust storm and one that hunkers down, saves power, and re-emerges when the sky clears.

What This Means for the Future of Space Science

Look, the ECF program isn’t flashy. There are no rockets on launchpads or astronauts waving from the space station. But these three projects represent a quiet shift in how we think about exploration. The old model was: send instructions, wait for confirmation, hope for the best. The new one is: build machines that can learn, adapt, and survive — because the universe isn’t going to wait for us.

Dr. Simoneau’s diagnostics will make heat shields safer. Dr. Yamada’s navigation algorithms will put spacecraft where they need to be. Dr. Voss’s planning systems will keep them alive when things go sideways. Together, they’re building the foundation for a future where spacecraft aren’t just tools — they’re partners. And that partnership might be the only way we ever get past the front porch of the solar system.