Today's Mars rovers typically drive at about 500 to 1,000 feet per hour — roughly the length of one to three football fields — and only for short stints at a time. 

NASA engineers say they could go so much longer and faster if they could only think more for themselves. 

A miniature prototype is already proving that point. On a recent test day, ERNEST, short for Exploration Rover for Navigating Extreme Sloped Terrain, rolled toward a pile of boulders that would stop Curiosity and Perseverance in their tracks. Instead of stalling or waiting for new instructions from controllers, it sized up the rocks, shifted its stance, and climbed straight over — no joystick from 200 million miles away necessary.

The robot, about four-feet long, looks a little like Wall-E or Johnny 5, with four mesh wheels in place of treads. Those wheels can lift and tilt, as if rising on tiptoes. ERNEST has already shown it can trundle about 16 miles across the Colorado desert in California. Engineers were hands-off during the excursion, mostly there to see how far the rover's new brain and body could take it. You can watch how it moves and navigates treacherous obstacles in the two videos below. 

The space agency's next-generation of rovers will sense hazards, pick their own paths, manage their power and health, and keep driving without waiting for detailed instructions from Earth. Micromanaging from mission control often means the current rovers sit idle between communication windows. But AI and onboard autonomy could reduce that downtime, allowing the machines to travel faster and farther into new frontiers.  

"Integrating artificial intelligence and autonomy software is the future of extraterrestrial rovers because it fundamentally overcomes the communication delays and communication resource constraints associated with space exploration," Ashish Goel, a research technologist at NASA's Jet Propulsion Laboratory, told Mashable. "We have already seen that the Perseverance rover is able to drive significantly more distance than the Curiosity rover, despite having a similar mechanical design, primarily because of its ability to think and drive, and better autonomous driving capabilities in general."

NASA has not announced a mission that would use ERNEST directly, but the agency has been studying concepts for future long-range lunar rovers that would require much greater independence.

One concept, known as Endurance, envisions a rover capable of traveling more than 1,200 miles on the moon over a span of four years, collecting and delivering samples across the lunar surface. To hit that goal, the rover would need to cover 2.5 to 4 miles per Earth day — about what Perseverance does in a year — and keep moving even when Earth can't send new instructions.

That scale of driving creates three big problems for human operators. First, communication takes time and literal bandwidth. A rover can't wait around for engineers to approve every turn of the wheel if it needs to cross a continent's worth of craters. Second, the schedule can't afford daily hand‑holding. Mission planners talk about uplinks, the batches of commands NASA sends as radio signals to a rover, only every couple of weeks, not every day. Third, the rover must survive and stay productive through harsh swings in light and temperature, including long two-week stretches of lunar night.

To handle that, the team behind Endurance argues for true onboard autonomy, not just smarter cruise control, according to a paper from the 2025 IEEE Aerospace Conference. The rover must control its own route, power use, thermal limits, and health checks. It has to pick paths that meet science goals while steering clear of boulders, craters, and sand traps. When something goes wrong, it needs to diagnose at least some problems on its own.

"The frequency of anomalies," the authors wrote, "has to be sufficiently low to maximize the mean-distance between interruptions, i.e., the distance traversed before the rover has to stop and call home for help."

That's where AI and ERNEST come in. Engineers first built even smaller, half‑scale rovers and ran them through a sandbox full of artificial moondust. They tried 11 different ways to control the suspension before they arrived at the design. The prototype can squat, lean, or walk over obstacles and spread its weight when the ground looks soft.

To replace a human operator, the team turned to reinforcement learning, a type of AI where a robot learns by trial and error. JPL's Dynamics and Real‑Time Simulation Lab built a virtual world that mimics ERNEST's behavior on different slopes, surfaces, and rock fields. Engineers fed the simulator the data from real hardware tests, then virtually drove for thousands of hours.

After months of digital practice, they moved the new driving "brain" into the real rover and sent it into the lab's Mars Yard, a sort of obstacle course for extraterrestrial robots. There, ERNEST navigated sand ripples, rubble piles, steps, and steep slopes. It chose when to lift a wheel, when to crab crawl sideways, and when to pick a safer way around trouble.

But autonomy has to reach far beyond clever footwork. A rover needs sharp perception to navigate through some of the trickiest lighting conditions in the solar system. On the moon, for instance, sunlight can blast one side of a scene and plunge the rest into deep shadow. Dust and rock don't scatter light the way air does on Earth, so cameras can see both blinding glare and pitch dark in the same frame. That mix makes it hard to spot hazards and estimate distance.

To address those challenges, engineers want to give future rovers sharper "eyes" — better cameras, headlights, and laser 3D mapping — so the onboard AI has enough information to make informed steering decisions. 

"As more capable computers become available for space missions, we can leverage the advancements in AI and autonomy to go farther, faster, and into more challenging environments, to unlock more scientific insights from our future planetary rovers," Goel said.