Man's Best Friend Goes it Alone
Continued...

Many ideas were passed back and forth among the group - which included engineering faculty members George Flowers and Tony Overfelt - and each was considered for its utility, practicality and potential of interest for governmental research and development funding. When the idea was presented to use GPS and autonomous guidance with canine training, the CDRI was more than interested; as a program that had started with bomb detection, faculty and staff understood the need for a method of handling detection canines remotely.

Bevly had been invited to sit in on the discussion and decided to use those topics to finalize his ONR proposal. He amended his plan and was awarded funding based on an idea to research the use of dogs for security assistance without the immediate presence of a handler. The project would use GPS and a microcontroller to automatically guide trained dogs to specific locations.

"The technology was much like that used with unmanned ground vehicles, but the modeling of the system and the design of a unique control algorithm for a dog would be the focus of the research," said Bevly. "This project was similar to others our lab had worked on, but now our focus would be a canine instead of a vehicle.

" With funding in hand, Bevly immediately began looking for graduate students who could assist with the project. Because the plan was to actually build a pack for the dog to wear, as well as develop a software program, he began the hunt for an electrical and computer engineering graduate student who would be up to the task. However, it was a computer science and software engineering faculty member whose research caught his eye, leading him to then doctoral student Win Britt.

"I went to faculty member Gerry Dozier to see about using neural networks to guide the dogs," Bevly said. "Then Win came on board with a plan to use an artificial intelligence method." \

As a budding computer scientist, Britt jumped right to artificial intelligence as a means for guiding the dog.

"I just began by asking the question, ‘How can we model and control a dog?'" said Britt.

The plan was simple. Or at least simple to explain.

The team would develop a suite of models and algorithms that could autonomously direct a canine to a specified point, using GPS coordinates. The canine, would wear a device that would utilize the programmed directions to lead him to that position.

Britt and his team spent the better part of three years making it all come together.

He focused specifically on the modeling and software systems that would autonomously control the canine and worked side-by-side with mechanical engineering graduate student Jeffrey Miller, who was responsible for the GPSrelated aspects of the program.

"To be able to automate the remote control of a canine, we needed to attach sensors to the canine that would give the computer knowledge about the canine's position and heading in reference to given way points," said Miller. "My job was to incorporate GPS with inertial sensors to give that information." The team also included two undergraduate research assistants, computer science's William Lyles and electrical engineering's Stephan Henning.

"What became really interesting was how to build the controller," Britt said. "It was fascinating what we could learn from those little sensors and how we could use the data that they retrieved."

They worked countless hours putting together a canine vest that would receive the signals to direct the dog. Through the use of tones and vibrations, the team discovered a system that could direct the dog to go left or right, forward, come to a stop or return to its base and trainer.

But they couldn't just place the outfit on a dog without trial.

"Oh, yes," Britt said with a chuckle, "we definitely spent some time testing the equipment on ourselves first."

After numerous rounds of lab testing, the team decided it was time to take their equipment out into the field. For this stage of testing, the team chose a 4-year-old yellow lab named Major from the Canine Detection Training Center, part of the CRDI.

Major was already well trained in "blind retrieves," a method often used by hunters to retrieve game fowl, such as duck, pheasant or quail. In a blind retrieve, a hunter uses sounds or hand signals to direct the canine to the desired point to retrieve its quarry. This way, the dog isn't confused if he loses his line of sight. This training made Major a perfect choice for the project.

"Major already knew the mechanics that went along with being directed to a specific goal," said Britt. "That made our job easier in that all we had to do was change the source of the stimuli he was used to."

Once he adjusted to the sensor vest, Major was ready to test his skills at finding his hidden "dummies."

Britt arranged for two different types of The first, which he calls the "fair test," involves a situation where a person with remote access can accurately provide the same directions as the control computer. The second, the "unfair test," involves placing the dog in a setting where the computer must give all the directions, such as a situation in which a building or similar distraction blocks the trainer's line of sight.

For the most part, the field trials went smoothly.

"Occasionally, we'd forget that the dog didn't really have a sense that he was carrying one-of-a-kind equipment in a vest on his back," Britt said. "So, he'd run too close to the side of a building and clip the electronics pack, or he'd slide through a mud puddle and soak the equipment."

<< Page 1    Page 3 >>