Defense Technology International - January/February 2008 - (Page 32) DISPATCHES GLOBAL battlefield problems. Two years ago, there was no mention of robotic convoys in briefings on the Army’s $300billion Future Combat Systems family of manned and unmanned ground vehicles. But last October at the Assn. of the U.S. Army trade show in Washington, military and industry officials highlighted robot convoys as a potential adjunct and 2005, and Urban Challenge this year. Perhaps the two most important developments—both vital to CAST—are Lidar sensors and the decision-making software that allows a robot to pick between the countless choices it faces on a constantly changing road environment. (CAST uses the Lidar sensors from Germany’s Sick AG that have proved popuwaypoints to the followers. But since GPS can be degraded by distance or jammed, CAST relies on GPS signals just 1% of the time. “GPS is ultimately a backstop, but not a very good one,” says Peter Jarvis, a PercepTek test driver. The sensors’ capabilities overlap considerably. CAST is essentially a collection of modulating behaviors, each relying on a slightly di erent combination of sensors, all intended to keep a bunch of trucks in line and moving without hitting each other or other vehicles. “There are multiple behaviors, each with its own algorithm,” Jarvis says. The “road follower” behavior, for instance, senses the shape and direction of the road and keeps the truck in its lane. The “vehicle follower” behavior keeps track of the lead truck and tags along right behind it. Both behaviors function to keep a truck moving along the road in generally the right direction. The “confidence” of each behavior is indicated by a light on the CAST console. “If the road follower has low confidence, it cedes to the vehicle follower.” This nimble back-and-forth between complementary modes ensures that CAST copes with rapidly changing conditions. “The art of it is knowing when to switch,” says Chris Hall of PercepTek. He says refining the software “arbiter” that decides between modes requires “testing, data and common sense.” After the A.P. Hill experiment, the CAST team journeys to Tardec for a demonstration, then goes to Denver for tweaks before launching another round of experiments. One lesson learned at A.P. Hill, according to Jarvis, is that automatic braking needs to be smoother. On the other hand, the team discovered that CAST was better than human drivers at detecting pop-up obstacles—an animal darting across the road, for instance—and screeching to a halt. Whether soldiers are ready to trust a robotic driver, however sophisticated, is still an open question. Asisat Animashaun, an investigator from the Army Research Laboratory in Aberdeen, Md., was at A.P. Hill to study the psychological e ects of mixing people and robots on long-haul supply runs. Her results are pending, but one thing is clear: “As the system becomes more robust, trust increases,” she says. “We’re building trust through exposure.” Workload and fatigue are considerations, but Animashaun says three of four participants in one test group showed an average 25% improvement in situational awareness when they were able to take their hands o the wheel. I www.aviationweek.com/dti The CAST interface is limited to one console with just four buttons (above). A PercepTek driver takes his hands off the wheel (below) and lets the CAST robotic autopilot take over. DAVID AXE/DEFENSE TECHNOLOGY INTERNATIONAL PHOTOS to FCS’s core combat capabilities. Maj. Gen. Charles Cartwright, FCS program manager, says that the “Mule” armed robot includes “follower” capabilities that “could be spun out into future robotic convoys.” As with CAST, a Mule-derived robo-convoy would still require a human leader, at least initially. “But it goes faster if [the trucks] follow like baby ducks.” That huge boost in confidence reflects the rapid advancements in robotics since 2004, many of them spurred by Darpa’s Grand Challenge races in 2004 32 lar on the Darpa Challenge racers.) CAST vehicles rely on a range of redundant sensors. The Lidar ones, which scan an environment with lasers to create a 3D model that can be interpreted by a computer, work great in clear weather, but have trouble with dust. Optical cameras are simpler and more reliable, but don’t work as well in rain and darkness. The radar, a spin-off of civilian Active Cruise Control technology, is good for detecting other vehicles. CAST boasts a GPS, too, so the lead vehicle can feed DEFENSE TECHNOLOGY INTERNATIONAL JANUARY/FEBRUARY 2008 http://www.aviationweek.com/dti
For optimal viewing of this digital publication, please enable JavaScript and then refresh the page. If you would like to try to load the digital publication without using Flash Player detection, please click here.