Out on the rocky horizon, the robot has stopped dead in its tracks. "Uh, Dave, I got a big problem out here," a voice crackles over the radio.

"OK," David Wettergreen replies carefully, peering off in the direction of the machine. "Define 'big.'"

"Big" turns out to be a new part for the robot that doesn't quite fit and so prevents the robot's cameras—its eyes—from turning properly. Back at the laboratory, this would be a quick fix, but the robot, Wettergreen, three geologists, two software engineers, two sociologists, an electrical engineer, a mechanical engineer, and a biologist are all out in the middle of Chile's vast Atacama Desert, [see map] many hours' drive from civilization.

As he strides off to investigate, you get the sense Wettergreen's enjoying himself. For the better part of an hour, he and two colleagues will wrestle with the aberrant part [see photo, "All in a Day's Work"]. The tedious work produces the standard amount of swearing, but they also joke—when one of them wields his drill stick-'em-up style, Wettergreen gamely throws up his hands. Finally they conclude that the job would be better handled back at the base camp.

Wettergreen, an associate research professor at Carnegie Mellon University's Robotics Institute, in Pittsburgh, and his team have been roughing it here in the Atacama since August, and they'll remain until November, just as South America's spring gives way to summer. They've come to test out new concepts and designs for the next generation of planetary rover, because this place, more than any other on Earth, approximates the barren, arid rockiness of the Red Planet. Testing the rover means pushing the technology to its limits, and sometimes beyond. The robot is so unusual and so new that breaking down is, for now, anyway, what it's supposed to do. "A hundred percent success means we're not really trying hard enough," Wettergreen says.

It isn't the most elegant-looking machine ever built. Weighing in at 180 kilograms, the rover, dubbed Zoë, looks something like a motorized, overgrown ice cream cart. But it is beautiful in the one way that really matters to planetary scientists: unlike all the rovers built thus far, Zoë can roam autonomously [see photo, "Autopilot"]. Its lesser rover cousins still need human drivers back on Earth to issue steady streams of commands that enable the robots to pick their way gingerly among the boulders, slopes, and ridges that constantly threaten to trap or upend them; one false move could terminate an interplanetary mission costing hundreds of millions of dollars [see "Meanwhile, Back on Mars..."]. Zoë is smarter: it can sense, among other things, when it's on an incline of more than 30 degrees or nearing a too-precipitous drop-off; in such situations, it is programmed to seek an easier route.

The rover can even make some rudimentary decisions about what terrain to explore. In a set of experiments conducted in Chile, Zoë successfully determined which tests to run at a given location. It started by taking an initial image of the spot; based on the density and types of rocks it was seeing, it calculated the probability of finding life there. When it figured the probability to be high enough, it ran through a sequence of tests to look for chlorophyll; when it detected chlorophyll, it went on to check for carbohydrates, proteins, and other signs of life.

The calculations are all done with hardware that is much less powerful than your typical desktop PC. All of Zoë's six computers are off-the-shelf products designed for factory automation, telecommunications, and other industrial systems. Two 2.4-gigahertz Intel Pentium 4 processors control the robot's navigation and autonomy. A computer based on an Advanced Micro Devices Celeron chip estimates the robot's position, while an Intel Pentium microprocessor looks after power management. Two computers control the robot's motion.

The idea is that eventually, scientists back on Earth won't need to send step-by-step instructions to the robot; if it spots a rock of particular interest, it will just mosey on over and investigate, instead of waiting for a human to tell it what to do. A fully autonomous rover is still a ways off, Wettergreen says. In addition to making decisions about what instruments to deploy or what tests to run, such a machine also will need to consider the overall picture. If I veer off in this direction to explore that patch of ground, how much power will I consume? How long can I spend on the task without compromising other activities? And so on.

"The simple things will happen first, maybe even in the next generation of rovers," Wettergreen says. "But it will be a long time before we see complex behaviors showing up in flight systems."

Wettergreen and his team built Zoë with a US $3.9 million grant from NASA, and they've come to the Atacama to see how well the rover copes with the kind of terrain its successors will find on Mars. At Zoë's top speed of 1.2 meters per second, its human "wranglers," who trail the robot at all times for safety's sake, have trouble keeping up. The rover's wide, boxy body sits low to the ground on fat mountain-bike tires; its shiny back of solar panels feeds two racks of lithium-polymer batteries below. Perched atop a long metal stalk of a neck are three high-resolution color digital cameras, used to look at terrain off in the distance; individual images from these cameras can also be stitched together to build panoramic views. Further down, two wide-angle navigation cameras look a few meters in front of the robot to detect obstacles.

Ungainly as the robot looks, Zoë in motion has a certain gracefulness. Each wheel is driven independently, while the front and rear axles, which attach to a central "spine," pivot passively—going over bumps, the robot's chassis appears to undulate. This arrangement allows the wheels to remain on the ground at all times and gives the machine a tight turning radius. But controlling the drive isn't simple, Wettergreen says. "It's not just a matter of giving the wheels the right velocities. You have to be a little bit predictive." Engineers at the Robotics Institute, where the robot was designed and built, added a mechanical linkage that averages the rolls of the front and rear axles and the height of any obstacles the wheels are going over and then distributes the load accordingly.

Zoë, whose name means "life" in Greek, is the prototype of a vehicle that will likely rove Mars in the not-too-distant future, hunting for evidence that some kind of microbial life flourished in the planet's warmer, watery past, as well as signs that some of it might have held on to the present day. The most important instrument in its suite is a fluorescence imager that exploits the fact that certain substances fluoresce when exposed to light at certain frequencies. It will inspect rocks and dirt for the presence of chlorophyll, lipids, carbohydrates, proteins, and DNA—the chemical signatures of life. Chlorophyll glows naturally when excited; the others do so only when treated with special dyes.

In the lab, a fluorescence imager would excite the sample using high-power lasers, each tuned to a particular frequency. But the robot's unit has to be compact and low-power, so it uses a xenon flashlamp, which produces a 10-millisecond, 1000-watt burst of full-spectrum light, explains Shmuel Weinstein, a biologist at Carnegie Mellon who helped build the imager. The light passes through a filter wheel, which lets through just the frequencies that a sample might emit when excited, and then into a fiber-optic bundle, which directs the light to the sample below. The imager also has a system for automatically applying the dyes, water, and a mild acid (distilled vinegar, actually, which breaks down cell walls and lets the dye penetrate) to the sample. A charged-coupled-device (CCD) camera takes pictures of the sample, first without the dye and then with it; the robot then compares the images to see where the sample is fluorescing.

Even after Zoë has gathered its pictures, the scientists carefully reexamine the spot by hand, collecting rock samples and, if necessary, running a portable spectrometer over the ground. Later, they'll compare the robot's data with their own. This process, known as ground truthing, will tell them whether the robot can be trusted. On Mars, researchers won't have that chance, so it's better to know now.