Keeping dogs happy and healthy with an AI dog trainer

Why this Silicon Valley-based start-up is putting automated training in reach of every animal shelter and dog owner

ic_share_24px Created with Sketch.
Close Icon Created with Sketch.

According to the ASPCA, some 3.3 million dogs enter U.S. animal shelters every year. Of those, only 1.6 million get adopted.

And of those, 670,000 are euthanized.

Even among the dogs that get adopted, many get sent back to the shelter to be “re-homed,” or placed, with another family, if they are lucky. The most common reason that adoptees get returned: behavioral problems, because training dogs is difficult and expensive and shelters don’t typically have the resources to provide enough of it. Training a dog costs more than $100 per class, according to the ASPCA. With a minimum of six classes required to teach basic commands, that makes training the single biggest expense of owning a dog.

Silicon Valley-based start-up Companion Labs is changing that by putting automated training in reach of every animal shelter and dog owner. Training that could not only save more dogs, but also has the potential to save human lives. It all comes down to a box, smaller than a minifridge, with artificial intelligence on board

CEO John Honchariw founded Companion Labs. A life-long dog lover, ex-Googler Honchariw had seen a way to do good for dogs, and their owners and handlers, by applying his expertise in engineering and business. He found a willing partner in the San Francisco ASPCA, co-locating his operation there, so he could iterate more quickly. After three years of design, building, and testing there, the company was ready to begin to realize Honchariw’s vision to help dogs the world over. “We’ve created this simple, beautiful box that can understand and interact with your dog,” he said. The company began shipping its CompanionPro unit in May 2020. “If you put those boxes in shelters, it can help animals get adopted, stay adopted, and help them be better companions.”

The Ever-Present Companion

Honchariw’s “beautiful box” incorporates a treat dispenser, a camera, speakers, and a Coral System-on-Module. The module, Honchariw says, is production-ready with no modifications, speeding time to market.

Powered by what may be the world’s largest image database of dog behavior, the CompanionPro uses the onboard camera to watch when a dog exhibits a desired behavior, such as obeying orders like “sit” or “stay,” and automatically dispenses a treat.

The key to its responsiveness is machine learning algorithms that run locally, reacting to visual cues faster than any cloud-based system, and more consistently than a human trainer over many hours.

The company chose Coral through a competitive process, Honchariw says. He cites three reasons for choosing Coral for local AI. “One, the performance is phenomenal.” The Coral system was less expensive than any other system the team evaluated, yet ran at a higher frame. That allowed it to run Companion Labs’s proprietary algorithms faster than the competition, according to Honchariw.

“Two, it was super easy to use off-the-shelf, in a variety of ways. Not only is it a stand-alone research platform with the full developer board, but also you can take the system-on-module right off that developer board and put it onto something production, which is super cool. And three, the team’s just phenomenal. I have a ton of confidence that Google is going to continue to do well there.”

Sniffing out Disease

Honchariw’s dream of using technology to save dogs, came before the global pandemic claimed hundreds of thousands of human lives. Now he’s working on ways to help dogs save humans. Dogs have long been used to sniff out drugs and bombs, but only more recently used to detect disease.

“Dogs can distinguish the T-shirts of people with Parkinson’s from those without the disease,” points out Maria Goodavage, author of Doctor Dogs: How Our Best Friends Are Becoming Our Best Medicine (Dutton, 2019). “They can suss out socks that have been worn by children with malaria. In Vancouver, dogs walk the halls of hospitals sniffing for the dangerous superbug C. difficile to keep it from spreading and harming vulnerable populations.” It’s possible dogs could also detect COVID-19.

Honchariw is working to find out, with a new partner, the National Institute of Canine Service and Training (NICST), also based in the Bay Area. Just as Companion Labs realized the potential to use dogs to sniff out COVID, so did NICST, whose initial focus was detecting diabetic episodes and has now also broadened its focus. According to founder and CEO, Mark Ruefenacht, NICST began testing whether rats can smell COVID-19 infections in June, as a precursor to working with dogs.

Happy Dogs, Healthy Humans

If dogs are able to detect COVID-19, they will need to be trained in large numbers to make a real difference in keeping disease out of workplaces, schools, and other public places. That’s where Honchariw sees Companion Labs coming in. He’d heard of research into whether dogs could identify COVID infections in Europe, learned of NICST through a colleague, and saw an opportunity to partner with the local effort to provide the scale of training lacking elsewhere.

“Our intention is to go work with a variety of agencies that would be in a position to scale this nationally or statewide,” Honchariw says. “There are tens of thousands of working dogs right now.” What if they could all learn to sniff out disease in real-time so that sick workers, travelers, and others could be sent home to stop the spread of disease? CompanionPro could make it happen, Honchariw says.

Ruefenacht agrees. “Homeland Security itself would have a massive pool of dogs, already with handlers, that would just need to have an added skill, rather than be trained from scratch.” He also believes Companion Labs could give dogs and their trainers the critical edge they need to learn to detect specific diseases more rapidly than any testing technology.

All of which would solidify dogs’ position as humankind’s best friend, with help from local AI.