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

Overall, the project was successful. The computer correctly issued directional commands 99 percent of the time.

However, Major's response to left and right directional commands was only accurate about 80 percent of the time. "You have to understand that dogs see left and right a little differently than we do," Britt said. "You could give perfect directions, but you didn't always get the ‘left' or ‘right' that you wanted."

Britt used this research project as the topic for his doctoral dissertation.

"The most significant thing about this research is that no one has ever autonomously controlled a canine in the manner we demonstrated," he said. "I guess you could call Major the first autonomous remote-controlled dog."

In August, Britt graduated with a doctorate in computer science and software engineering. His research project will continue under his former team.

"We've asked for an additional year on the funding from the ONR," said Bevly. "There won't be any more funding through that particular program project, but we can tack another year of work onto the resources we already have and then can search for further funding from other sources."

Miller and the undergraduate researchers will take the project to the next level, working towards a better understanding of the directional issues Britt encountered.

"Now that we can accurately locate the position and orientation of the canine, the question becomes what algorithms will be effective in getting the canine where I want it?" Miller said. "Other companies are showing interest in the work now as well, and that gives us the opportunity to expand from our initial goals."

While Britt focused on the use of GPS and machine learning to autonomously control Major, Miller plans to add the use of inertial sensors to the vest and investigate other control strategies. GPS gives position, velocity and course measurements with excellent accuracy, as long as there are enough satellites in view and the location of the satellites is conducive to good estimates.

However, GPS measurements are taken at relatively low rates and can be non-existent if the canine moves into areas that have low satellite visibility. So, inertial sensors and magnetometers can be used along with GPS to help out. Low grade inertial sensors, such as gyroscopes and accelerometers, give measurements at high rates and continue to do so even when the canine is indoors and out of satellite view. But, due to measurement errors, their output is prone to drift from the true solution.

"GPS can be used to estimate those gyroscope and accelerometer drift factors when we have a good GPS lock," said Miller. "When we lose GPS, our inertial sensors can still give us a decent estimate of position until we regain GPS."

One thing the team discovered during field testing was that Major actually started to learn the algorithms and understand what the computer wanted. Instead of each direction being a surprise, the dog could learn what the computer was asking even before it directed. As he learned what the algorithms were conveying, Major's performance level increased. Eventually, the team would like to be able to allow the algorithms to train the dog.

"This activity requires a unique blending of technologies from mechanical engineering, software engineering and applied behavioral science of CDRI," Waggoner said. "As is often the case, this effort also resulted in important basic science and instrumentation developments that have potential practical benefits other than those for military or homeland security, such as in employing instrumented dogs in assisting persons with various disabilities. Such activities are consistent with the CDRI's mission of enhancing working dog technology for the protection of the people and critical infrastructure of the nation."

The autonomous canine system's hardware is surprisingly small. The smaller box (right) is a custom-made lithiumpolymer battery, developed by undergraduate researcher Stephan Henning. The larger black box is the system's microcomputer - built and programmed by the team - while the smaller orange box contains accelerometers, gyroscopes and magnetometers.

<< Page 2