Blogs

From Ars Technica:

[On the Antarctic ice], an ever-present danger is posed by crevasses: rents in the ice sheet up to 60 meters deep and 9 meters across—big enough, at their widest, to swallow a supply-towing tractor whole. New crevasses are formed by movements in the ice sheet, but they can be hidden from view by fresh snowfall. That snow can bridge the gap, but it's incapable of supporting a supply vehicle's weight.

For about 20 years, the solution to this problem has been to manually survey for crevasses with ground-penetrating radar (GPR). A radar antenna, often housed in an inner tube, is pushed along on a 6-meter boom by a PistenBully or some such vehicular beast of burden at the head of the traverse. Alongside the driver rides a GPR operator who monitors real-time scrolling radar images, looking for the tell-tale signature of a crevasse. "If a crevasse is detected, the operator has about two seconds to stop the tractor before it crosses," Laura Ray, Professor of Engineering at Dartmouth, explained to Ars. If a crevasse is found, it is probed to see if it is passable. If it is not, the crevasse can be blasted and filled, or an alternative route can be found.

But approached at a shallow angle, crevasses can go undetected, and unusual GPR signatures can make crevasses difficult to identify. Unseen, a crevasse is as dangerous to a GPR survey vehicle as it is to a supply tractor. James H. Lever, a mechanical engineer at the US Army's Cold Regions Research and Engineering Laboratory, has conducted manual GPR surveys and photographed the dangers: collapsed snow bridges and (thankfully) minor incidents of vehicles partly caught in crevasses. "These incidents caused no damage and only minor delays, but they left a lasting impression," reads the accompanying caption in a new research paper published in the Journal of Field Robotics. Interpreting these accounts, one is left with the impression that the process of manual surveying would be extremely tedious were it not for the occasional moments of considerable risk. The process is repetitive and stressful, or, in other words, ripe for automation.

A robot to the rescue

The paper in question is titled "Autonomous GPR Surveys using the Polar Rover Yeti," which neatly summarizes a solution to the problem. A robot with a GPR antenna in tow (which at about 80kg is light enough to cross snow bridges) is a tremendous ally when it comes to traversing ice sheets. Following jaunts along the Greenland Inland Traverse to the NSF's Summit Station and from the McMurdo Station to the South Pole, Yeti is claimed to be the first robot proven in the field to be capable of identifying crevasses hidden under the snow.

Roughly a meter wide, the GPS-equipped, four-wheel drive robot follows preprogrammed routes to survey areas for hidden fissures. For lightweight vehicles, wheels with low-pressure tires are thought to offer decreased turning resistance compared to caterpillar tracks. Perhaps surprising for a vehicle that operates at temperatures as low as -30°C, Yeti is powered by six lithium-ion batteries capable of powering the in-wheel motors. Yeti trundles over snow at a speed of 2.2m/s for between 2.2 and 9.3 hours. Pleasingly, researchers resorted to chemical hand-warmers to keep the batteries and GPR controller warm, the latter being even more sensitive to cold temperatures.

Though Yeti can be driven manually, automation is necessary for accurate surveying. To master the tricky problem of approach angles, Yeti follows looping paths, outward and back to a central point, as if describing the outlines of flower petals. "A grid pattern may be used to do a broad survey to locate possible crevasses. A rosette survey helps determine the direction of the crevasse and provides us with data from multiple approach angles," explains Ray, who co-authored the paper. "When a crevasse is crossed perpendicular to its length, it is most easy to see on a radar image. However, when it is approached at a shallow angle—which happens routinely because crevasse directions are not known in advance—its signature lacks characteristics (parabolic reflections) that make it easy to detect by human eye. The rosette pattern provides us with multiple crossing angles so that we have data from many characteristic crossing angles to develop our classification algorithms."

Because identified crevasses are positioned relative to Yeti, it's imperative that Yeti's GPS positioning is accurate. In the field, Yeti typically acquires 10 to 12 satellites for a position accurate within ±3 meters, which the authors describe as "acceptable" for locating crevasses.

Though unrelenting in its search for hazards below ground, Yeti is completely blind to those above. The relative lack of surface obstructions on the ice sheets makes the cost of obstacle-recognition systems hard to justify. In the rare event that Yeti's path is blocked (unusually soft snow is a problem), the team takes manual control.

Though Yeti relies on a human operator to identify possible crevasses in the data, it is hoped that the database of signatures gathered from encounters with hundreds of crevasses will help to develop algorithms for autonomous crevasse detection in the future. There's hope, too, that future robots will be able to detect and avoid obstacles in their path.

Yeti is a multitasker

It turns out that Yeti is good for more than mapping crevasses. In recent months, Yeti mapped the ice caves on the approaches of Antarctic volcano Mount Erebus. It's thought that, thanks to volcanic outgassing and the warmer conditions inside the caves, they may play host to undiscovered microbial life. Yeti has also mapped the abandoned 1950s Old Pole research facility, now buried under 3-10 meters of snow. An accident in 2010 had seen a tractor "accidentally discover" a top hat, a false attic built to reduce the weight of snow while the facility was in use. The remaining top hats were blasted and naturally in-filled with snow. A subsequent Yeti survey assessed the success of the operation to determine if the Old Pole site was safe to reuse. In part a learning exercise, Yeti managed to detect a separate unknown outbuilding that prior surveys failed to pick up.

Asking Ray what would make the shopping list for Yeti II, new wheels are top of the list. "Oddly enough, we probably would not consider more sophisticated sensing systems, at least not in the environments in which we operate Yeti," she tells Ars. "The most significant change might be in the wheels. We used off-the-shelf aluminum hubs with rubber ATV tires. Custom wheels could reduce mass significantly, and increasing their width would reduce ground pressure. Reduced ground pressure means the robot will sink less, and thus resistance is reduced."

Yeti will return to the field this year to check the Old Pole site one more time, as will its predecessor, the atmosphere-sampling Cool Robot. Polar science looks set to follow space science in putting robots on the front line of research in the field.

Journal of Field Robotics, 2013. DOI: 10.1002/rob.21445 (About DOIs).