Some of the most scientifically interesting sites on Mars are also some of the hardest to get to. Layered terrain exposed on the cliff faces of deep canyons. Gullies etched into the sides of ancient craters - possible evidence of the presence of liquid water on modern-day Mars. These are some of the locales that scientists would like to explore.

But to the rovers that have been sent to Mars so far - Sojourner, Spirit and Opportunity, and even the Mars Science Laboratory, slated for launch in 2009 at the earliest - sites like these are inaccessible. They're simply too dangerous.

A group of engineers at NASA's Jet Propulsion Laboratory (JPL) in Pasadena, Calif., is exploring a novel solution to that problem. Cliffbot is a small wheeled rover that works as part of a three-rover team. It is attached by tethers to two other rovers, "anchorbots," a configuration that enables Cliffbot to navigate terrain as steep as 85 degrees.

Before Cliffbot can be sent into action, the two anchorbots must first be secured on the edge of a cliff. Cliffbot is then tethered to the anchorbots and lowered down the cliff face to perform scientific experiments and collect samples.

The approach, says JPL's Paulo Younse, the mechanical lead for the Cliffbot system, is "modeled after tether-aided human climbing." The two anchorbots contain winches that spool out or reel in cables, so Cliffbot can position itself precisely and maintain its stability.

The winch mechanisms are modified fishing reels; the braided cables are a synthetic material called Spectra, also used in deep-sea fishing, that is compact, lightweight, resistant to corrosion and abrasion, and stronger than steel.

Cliffbot isn't passively suspended from the tethers, however. Each of its four wheels has a motor "about the size of your pinky," Younse explains.

"Instead of just lowering a robot around, you actually want to have it drive over terrain, and you can't necessarily do that by just tugging on the tethers. So the robot's actually driving" over the landscape. "The tethers just react to keep it stable."

On the tail end of the Cliffbot, where the tethers are attached, pitch and yaw sensors monitor the angle of the tethers. In addition, the anchorbots have sensors that track how much cable they've spooled out and how much force the robot is exerting on each tether. Using this sensor data, Younse says, "It's a simple little formula, just trigonometry, to figure out where the robot's at."

With ASTEP (Astrobiology Science and Technology for Exploring Planets) funding, Cliffbot was tested in the summer of 2007 as part of the AMASE (Artic Mars Analog Svalbard Expedition) project in Svalbard, Norway. In the 2007 field test, Cliffbot was outfitted with a robotic arm that contained a camera, a spectrometer, a micro-imager and a scoop for collecting soil and rock samples.

The AMASE team performed several successful runs with the system. In the longest of these, the robot climbed 13 meters (about 40 feet) down the face of a steep, rocky cliff.

"We did have one tipover," Younse said. But a bigger problem was with the batteries. The cold environment in Svalbard - it lies well above the Arctic Circle - "limits the lifetime of the batteries. We had to put them in our jackets to keep them warm."

Engineers also occasionally had to swap out batteries while the robot was suspended over the edge of the cliff. Younse and a colleague had to lower themselves, in harnesses, down to the robot's location, to execute the switch.

The AMASE team tested not only Cliffbot's technology, but its operation as well. They set up a scenario that mimicked the operation of a rover on Mars. The scientists and engineers giving the robot its instructions couldn't see the rover or the terrain over which it was traveling. They saw only the images the rover's cameras sent back to them.

They had to figure out "from that little black and white image what science, or what target looked interesting to be able to attempt a sample or be able to take a sensor reading from, and then figure out what kind of path to take," Younse said.

That process - giving the rover a set of commands; waiting while it traveled to its target, took images and other sensor readings, scooped up a rock and soil sample; studying the results; and developing a new set of commands - was much like the real process of controlling a robot on Mars. But it demanded patience.

The rover moves slowly, about 5 cm (2 inches) per second. That makes it easier to position the robot precisely, Younse explained. "And also, the slower you go, the more time the computer has to think, to analyze, and the safer it is, especially in a rocky environment, a very unpredictable environment."

"It was a little frustrating for the scientists," Younse said, "because they wanted to go right up to some of these positions and take a sample, but they had to wait for the robot to get there."

The biggest challenge the team faced, however, "was just getting the thing operational," Younse said. "We had to do a lot of scouting, more than we thought, to be able to find somewhere with a somewhat stable edge" to secure the anchorbots.

But at least in Svalbard, humans were available to perform the setup. Operating Cliffbot on Mars will present an even more daunting challenge. The rovers will have to scout for a safe location for the anchorbots, secure them, and set up the tethers without human assistance. "That's what will be necessary in the future," Younse says, "to be able to do it all autonomously."