Researchers at North Carolina State University have developed a computer system that can train a dog autonomously by using sensors to respond to a dog's body language. The technology could be used to more quickly train service dogs or as an add-on to make human-based training more efficient.
The technology includes a smart harness full of sensors that monitor the dog’s posture and body language. That information is collected and analyzed by a computer on the harness the size of a deck of cards. An algorithm triggers a beeping sound and the release of a treat from a dispenser when the desired behavior is achieved, like the dog going from standing to sitting.
“Because the technology integrates fundamental principles of animal learning into a computational system, we are confident it can be applied to a wide range of canine behaviors,” says Alper Bozkurt, an assistant professor of electrical and computer engineering at the university. “For example, it could be used to more quickly train service dogs. Ultimately, we think the technology will be used in conjunction with human-directed training.”
The real trick, the team discovered was timing the delivery of the treat to just the right moment. If the algorithm ran long enough to make sure that it was absolutely the correct posture, the treat was given too late to reinforce the action. If it was given too soon, it would often reward the wrong posture.
“That variation matters, because consistency is fundamentally important for all animal training,” said David Roberts, an assistant professor of computer science at NC State.
The team tested the technology by comparing it to human training. The algorithm rewarded the right behavior 96 percent of the time while the human trainers rewarded the right behavior 100 percent of the time. The only difference was in response time. The algorithm was very consistent while there was variation in the trainer response times.
While this was just a proof of concept, the researchers would like to start using the computer system to teach dogs to perform certain behaviors on cue and make human training more efficient.
“In the long term, we’re interested in using this approach to animal-computer interaction to allow dogs to ‘use’ computers,” Roberts says. “For example, allowing an explosive detection dog to safely and clearly mark when it detects components of a bomb, or allowing diabetic alert dogs to use their physical posture and behaviors to call for help.”