A scout with expertise in following trails through remote areas and rough terrain would be useful for a variety of tasks, from conducting search-and-rescue operations to assisting a hiker who has visual impairments.

If the scout also is a robot, it might be even more useful, able to work in otherwise inaccessible places, for longer periods of time and to carry out more dangerous tasks.

Creating such a device is the aim of Christopher Rasmussen, an assistant professor of computer and information sciences who is conducting groundbreaking research in computer vision and mobile robotics. He recently received the prestigious Faculty Early Career Development Award from the National Science Foundation (NSF).

“I feel very good about receiving the award,” Rasmussen says, “but I’m most excited to just get started on the work I proposed.”

Rasmussen will use the five-year, $500,000 grant to study autonomous and assistive trail following. He will investigate perceptual algorithms that could help mobile robots find and follow trails, as well as be embedded in portable devices to mitigate visual impairments or augment human tracking skills for search-and-rescue tasks, for example.

His primary area of research has been in computer vision for field robotics, or providing “sight” for robots in outdoor environments.

“There is a great divide between robots that roll down hallways delivering mail and robots that have to bounce around in the dirt,” Rasmussen, himself a hiking enthusiast, says.

The germ of his NSF proposal came from his experience as an adviser to a team from the California Institute of Technology in the Defense Advanced Research Project Agency’s (DARPA) Grand Challenge desert robot races in 2004 and 2005. His contribution was a software component for “road following,” or working out how to get a vehicle to drive itself along rough dirt and gravel roads.

Roads are one example of a more general class of paths that “show the way” across a landscape to robots and people who can recognize them and “ease the way” for those who follow them, Rasmussen says. These two functions place paths along a spectrum of distinctiveness, or how easy it is to see where to go next in a gross sense, and traversability, or how easy is it to actually negotiate a path that could be strewn with rocks or logs.

“A path, like a road, is engineered for high distinctiveness and traversability,” Rasmussen says. “I’m interested in a more difficult subset of paths, which I call trails, that are hard to see or follow. They may not even have been created purposefully—for example, people or animals walking along and leaving footprints or other signs.”

A successful trail-following system, or scout, would be able to differentiate diverse kinds of trails from their surroundings regardless of time of day, weather or season. It must accurately estimate the curvature of the trail and decide whether it forks or dead-ends, Rasmussen says. The scout also must be able to recognize when it has wandered off the trail and have some method for getting back on track.

Rasmussen adds that another aspect of the problem is to deduce the proper “line” within the trail for a rolling or walking robot to take “so it doesn’t slip, roll over or fall into a hole.” Dynamical issues are compounded by higher speeds and differing surface materials, such as sand, mud or rock. One approach Rasmussen plans to try is for the scout to observe humans as they follow trails, learning how they negotiate such obstacles as rocks and low-hanging branches, and then imitating them.

Rasmussen says a robust robotic scouting system would have a wide array of potential uses. Those include logistics, such as the resupply of remote camps; trail maintenance; wildlife study; and reconnaissance for border patrol or military scenarios.

While the major focus of the work is in robotics, Rasmussen says an important possible application is assistive.

“A scout system that can follow difficult trails on its own might be able to serve as a guide for visually impaired people,” he says. “Another example is expert human trackers who find and follow footprints for search-and-rescue or wildlife biology. What if the scout system could be carried by nonexperts and give them similar abilities?” Developing such capabilities will entail collaborations with researchers in wearable computing and human interfaces including haptics, which is the technology that pertains to the sense of touch.

Rasmussen says he became interested in artificial intelligence while an undergraduate at Harvard University. His doctoral adviser at Yale steered him toward computer vision as a specialty, and he began to work on field robotics problems while doing postdoctoral research at the National Institute of Standards and Technology. He joined the UD faculty in 2002.

“I’m fascinated by the quest to get computers to behave like people, to exhibit human-level intelligence at various tasks,” Rasmussen says. “To me, robots are the ultimate demonstration of artificial intelligence because they have to sense, plan and act over and over. When a robot is moving correctly and doing its thing, you get a feeling of wonderment watching it, like you’ve created something living.”

Rasmussen says that “exciting things are happening in this area right now,” mentioning that DARPA is gearing up for another Grand Challenge in 2007, this time in an urban setting.

“Out in the field is the frontier for robots,” he says. “You’ve got to leave the lab to tackle the most challenging problems.” He feels fortunate as well that this research will allow him to combine his love of programming and being outdoors.

The Early Career Development Award is one of the NSF’s highest honors for young faculty members. It recognizes and supports the early career activities of those teachers and scholars who are considered most likely to become the academic leaders of the future.