Somewhere in the Colorado Desert near Plaster City, California, a four-wheeled robot about the size of a large dog spent seven days in March doing something no NASA rover has done before: covering serious ground at serious speed. The Exploration Rover for Navigating Extreme Sloped Terrain — ERNEST, because NASA will never stop making acronyms — logged 16 miles over 37 hours of driving, topping out at 0.6 mph. That might not sound like much until you consider the competition. Curiosity and Perseverance, the two SUV-sized machines currently working on Mars, navigate at roughly a tenth of that pace.
An order of magnitude faster than anything NASA has put on another planet. And the rover did it largely on its own, using reinforcement learning algorithms to pick its way through sand ripples, rubble piles, steep slopes, and rock fields without waiting for instructions from a control room hundreds of millions of miles away.
Built for Terrain That Would Kill a Normal Rover
ERNEST is a Jet Propulsion Laboratory creation, four feet long with a rectangular sensor head perched atop a 4.5-foot mast, riding on four mesh wheels that give it an oddly insectile profile. The rover's party trick is an active suspension system with powered joints and gimbal articulation that lets it do things planetary rovers have never attempted. Each wheel can be lifted independently, allowing ERNEST to step over obstacles rather than trying to bulldoze through them. The system supports multiple locomotion gaits — the JPL team describes them as squirming, wheel-walking, and obstacle-climbing — and a clutch mechanism lets the rover switch between active and passive suspension modes depending on what the terrain demands.
All four wheels are independently steerable, giving ERNEST omnidirectional movement. It can sidestep. It can crab-walk. It can rotate in place. For a machine designed to explore crater rims, permanently shadowed lunar polar regions, and Martian canyon systems, that kind of agility is not optional — it is the entire point.
"You could do a science road trip across the Moon — or Mars — with this vehicle," said James Keane, a planetary scientist at JPL.
Teaching a Rover to Drive Itself
Speed without autonomy is useless when your rover is operating on a world where radio signals take between 4 and 24 minutes to arrive. ERNEST's navigation stack is built on reinforcement learning — the same broad family of techniques that taught game-playing AIs to beat human champions, adapted here for the considerably less forgiving environment of rocky planetary surfaces.
The training pipeline is genuinely interesting. JPL's Dynamics and Real-Time Simulation Laboratory built high-fidelity virtual environments where simulated versions of ERNEST could crash, flip, and get stuck thousands of times without bending any actual hardware. The lab runs extensive simulations, building up vast libraries of virtual driving experience. The simulation models are continuously refined against data collected from the physical rover's actual responses to terrain, closing the gap between virtual and real-world performance.
The result is a rover that does not just follow pre-planned waypoints but actively assesses terrain and chooses its own path. During the California desert trial, ERNEST was tested at dawn, dusk, and during full nighttime operations — conditions specifically chosen to simulate the extreme lighting environment near the lunar poles, where the sun hangs perpetually near the horizon and the landscape is a patchwork of long, harsh shadows. If you want to explore the permanently shadowed craters where water ice is believed to exist, your rover needs to operate in conditions where cameras struggle and hazards hide in darkness. ERNEST handled it.
From JPL Skunkworks to NASA Flagship Funding
The project's funding history tells its own story about how NASA develops risky technology. ERNEST began in 2022 as an internal JPL research and development effort — essentially a skunkworks project funded out of the laboratory's own discretionary budget. The hardware was completed by September 2024, and somewhere between concept and desert trial, the project caught the attention of people with bigger checkbooks.
ERNEST is now funded by NASA's Mars Exploration Program and the Exploration Science Strategy and Integration Office, both housed within the agency's Science Mission Directorate. That is a significant vote of confidence. Mars Exploration Program dollars are not handed out for speculative engineering exercises; they are allocated to technology that the agency believes has a credible path to flight hardware. The rover is managed by Caltech, which operates JPL for NASA.
Leading the effort are Issa Nesnas, a principal technologist and autonomy team lead at JPL, and Hari Nayar, a principal technologist and ERNEST team lead.
What Curiosity Keeps Teaching
While ERNEST was chewing through California desert, the rover it hopes to eventually succeed was still at work 140 million miles away. NASA's Curiosity rover, now deep into its exploration of Mount Sharp, has been ascending through a sequence of geological bands with distinctly different textures and tones. Recent operations around sols 4920 through 4926 had Curiosity transitioning from a dark-toned, rough-textured unit toward a lighter, smoother band above it.
The science return remains impressive. Curiosity deployed its Alpha Particle X-Ray Spectrometer for bedrock chemistry, its ChemCam laser for breakdown spectroscopy analysis, and its Mars Hand Lens Imager for close-up surface work across targets with names like Salto La Cascada, Puerto de Rosas, and Laguna San Rafael. Mastcam captured mosaics while Navcam scanned for dust devils — the full instrument suite working a geological contact zone that planetary scientists have been anticipating for years.
But Curiosity also demonstrated exactly why ERNEST exists. A delayed data downlink on sol 4923 forced the operations team to scrap planned drives and reschedule investigations — a routine frustration that underscores how communication latency constrains every aspect of current rover operations. Curiosity moves slowly not because its wheels cannot go faster, but because every drive segment must be carefully planned on Earth, uploaded, executed, and then verified before the next segment can be authorized. The rover creeps across Mars at a pace measured in tens of meters per day.
ERNEST's autonomy stack is designed to break that bottleneck. A rover that can assess terrain and navigate independently does not need to wait for a round-trip communication cycle before every move. It can cover the kind of distances that turn a geological survey into what Keane calls a "science road trip" — not a slow crawl between waypoints, but a sustained traverse across kilometers of unexplored terrain.
Why It Matters
NASA's current rover paradigm works, but it scales poorly. Curiosity has driven roughly 20 miles in nearly 14 years on Mars. ERNEST covered 16 miles in 37 hours. Even accounting for the differences between a controlled desert test and actual planetary operations — lower gravity, communication delays, instrument stops, the sheer hostility of space — the performance gap points toward a fundamentally different approach to surface exploration.
The lunar application is the nearer-term target. NASA and its commercial partners are planning missions to the lunar south pole, where water ice in permanently shadowed craters represents both a scientific prize and a potential resource for sustained human presence. Reaching those deposits requires a rover that can navigate steep slopes in near-total darkness, cover significant distances between landing sites and science targets, and operate autonomously when line-of-sight communication with Earth is blocked by crater walls. ERNEST's desert trials were explicitly designed to validate those capabilities.
The Mars application is longer-term but arguably more transformative. The next generation of Mars rovers could explore terrain that is currently off-limits — steep canyon walls, boulder-choked valleys, the flanks of volcanic shields — at speeds that make regional geological surveys practical rather than aspirational. A rover that can cover 16 miles in two days of driving rather than two years changes the fundamental calculus of what a surface mission can accomplish.
ERNEST is not a flight mission. It is a prototype, a technology demonstrator, a proof that the pieces — fast autonomous navigation, active suspension, reinforcement-learning-trained decision-making — can work together in the field. But the pieces are working. And the desert, for all its hostility, is considerably more forgiving than the Moon.
