Some days in the desert are better than others. Even for robots. Just a few weeks earlier, Zoë had a very bad day. The field team had to transport the robot by truck from Site D, on Chile's northern coast, to Site E, at Salar de Navidad to the south, about 400 km away over rough roads. They had the option of disassembling the robot, as they do when shipping it between Pittsburgh and Chile. But that requires several days of reassembly afterward, so instead, they loaded the fully assembled robot into the truck, secured it to the walls, placed wooden pallets underneath it to cushion the bumps, and hoped for the best. When they arrived at Site E, however, they discovered that the pallets had shaken loose and that Zoë's front axle had buckled and the rear axle had fractured completely. Several instruments, including the onboard spectrometer and the pan-tilt camera, also took a beating.

"My suspicion is that the wood underneath shifted and the robot basically started hopping up and down in the back of the truck," Wettergreen says. "The stresses on its axles would have been at least an order of magnitude greater than it would normally experience when it's driving."

Fortunately, the team had a spare set of axles in Pittsburgh. Jonak, the software engineer, was scheduled to travel to Chile anyway, so he loaded the spares into a ski bag and brought them down. Within the week, the robot was reassembled and nearly as good as new. Just to be sure, though, when it came time to move from Site E to Site F, the engineers seated the robot on a bed of rubber tires; no further mishap occurred.

Looking back, Wettergreen likes to think of it as an "extreme experiment." There's an entire discipline "where you test to the point of failure, basically," he says and gives a rueful smile. "So we did that."

Compared with the axle mishap, the various breakdowns, glitches, and bugs that occur on a daily basis seem fairly pedestrian. For example, Sols 2 and 3 at Site F bring a battery problem that necessitates temporarily swapping out the lithium-polymer packs for the spare lead-acid batteries; several software problems, which cause the robot to fail to execute part of its science plan and throw off the navigation cameras for a spell; a problem with the pan-tilt camera that causes the azimuth to slip and requires the entire camera unit to be dismounted, leaving Zoë decapitated for a time; and, perhaps most troubling of all, a lunch problem, caused by a shortage of bread back at camp. In place of the usual ham-and-cheese sandwiches, the cook sends a plastic bag of apples, oranges, and bananas; the field team grimly accepts the proffered fruits.

Often problems arise out of the best of intentions. Explaining the camera's software glitch, Jonak says, "There's kind of a black art to stereo calibration. You need to fiddle around with things so it's just so, and you're never really totally satisfied that you have the best system...you can always make these minor improvements. Unfortunately, that got us into trouble, because we tried to improve things, and we ended up knocking something out of whack."

In between repairs on Sols 2, 3, and 4, the robot fits in quite a bit of science and executes a few traverses of several kilometers. "We're traveling a lot farther than last year, things are running a lot more smoothly, and the [operations] are much more autonomous," Jonak says. "Despite these mechanical problems we've been having, we're still very happy with the rover's performance."

"You just have to embrace your filth." Wettergreen is leaning back in his chair at dinner, grinning, and trying in his own way to commiserate with a team member who's complaining about the impossibility of staying clean. It's the end of Sol 3, and we've all spent far too long coated with a fine layer of dust. It's in every pore and on every bit of hair, every garment. It's under our fingernails, in the creases behind our ears, in the folds of skin between our fingers.

Still, compared with the tents and sleeping bags at the previous two sites, the dormitories at Site F seem almost homey. The scientists still use sleeping bags, but there are real mattresses here to put them on, and the willful water heaters on occasion offer up a nice hot shower. There's no heat, though, and as the temperature plummets to below freezing at night, layering on every scrap of clothing you've brought along is the only way to keep warm.

But nobody would dream of missing the experience. "I like the desert," says Roxana Wales, a research scientist at NASA Ames Research Center, Moffett Field, Calif. "I find it very calming." The team members also get to experience things they won't soon forget. At Site D, near the coast, recalls Trey Smith, a Ph.D. student at Carnegie Mellon who worked on Zoë's software, "they have these salt fogs that come over the mountains and through the passes at sunset. Back in Pittsburgh when it's foggy, the air is still. But here the fog just races past you, like you're in a snow flurry. It's an amazing effect."

While Zoë may represent the next generation of planetary rover, what about the next next generation—and the one after that? "You're going to see a paradigm shift in planetary exploration soon," predicts James Dohm, a geologist at the University of Arizona, Tempe, and a member of the science team. "Think about it: you have vast areas that you need to cover. How are you going to do that?" Ground-hugging robots like Zoë can cover at most tens of kilometers a day. Aerial rovers, on the other hand, could travel much farther, Dohm says. "In the future we're going to see aerial robots as well as fleets of ground-based rovers. That will be a golden age for planetary exploration." Indeed, in December, NASA announced a new competition to design and build an autonomous aerial vehicle for conducting science investigations on planets and moons that have atmospheres. The competition, which ends in October 2007, carries a prize of $250 000.

Meanwhile, in January, project researchers convened in Chile one last time. The itinerary called for a quick visit to each of the six sites the robot had surveyed during the last three years. For the scientists who'd worked exclusively from Pittsburgh, it was the first time they actually got to see the places in person, rather than merely interpreting them through satellite imagery, science data, or photos. "The scientists had a great time, like kids in a candy store," Wettergreen says. "Mostly they wanted to get the big picture—which hill did the robot go over? Were we on that side or this side of the drainage? I think the visit confirmed everyone's expectations—they were happy they'd gotten it right."