It sounds like a joke: A team of researchers, one of whom worked at Google, is figuring out how to get dogs talking to humans.
But it’s not a joke. A team at Georgia Tech, led by Dr. Melody Jackson, is developing wearable technology that makes it easier for assistance dogs, and military dogs to communicate with their owners.
Down the road, as the technology advances, the hope is that dogs will be able to clearly communicate with all owners.
The project is called FIDO, which stands for “facilitating interactions for dogs with occupations.”
Jackson is a director at the school’s BrainLab. She’s also an assistance dog trainer. On her team is Thad Starner, who was a technical lead for Google Glass, and Clint Zeagler, another researcher who will be focused on textiles.
In her work at the BrainLab, she tries to understand brain signals, and uses sensors. She was also investigating computer interfaces for people with disabilities.
Working with Starner, she one day realised they could used wearable technology to get humans and dogs communicating. The way it works is a dog wears a vest with sensors that can send signals to the owner.
The research is still new, but Jackson says it’s showing signs of promise.
Here’s an edited transcript of a conversation we had with her about her research.
BI: Why don’t you just give some general background, you’ve been in this field a while, give us a little bit on your background and what you’re trying to work towards.
MJ: Essentially what happened is my lab partner Thad Starner, who is one of the pioneers of the wearable computing field, he and I have shared a lab since 2006 and we have collaborated on a lot of things, and we have a colleague from industrial design, Clint Zeagler, who is an expert in textiles and things like that. And it turns out that I’m also an assistance dog puppy raiser for Canines Companions for Independence, so I’ve trained dogs for assistance dog work since 1995. For some reason, it took us a while, but we finally made the connection that, “Wow, this wearable technology might be useful for assistance dog communication.” And I can give you a couple of scenarios.
One of my students, who has a guide dog, was travelling between classes one day, along a familiar route, just a sidewalk that he knew well, and the dog stopped still and wouldn’t move. That usually means there’s something in the way. So he got his collapsible cane out and he tried to feel around if to see if there was anything in the way and he couldn’t feel any obstruction at all. So he asked to go again, and the dog wouldn’t move. So he thought, “well, you know he’s just being a dog.” Though it’s rare, sometime the assistance dogs are just dogs, so he thought, “Maybe he’s sniffing something.” So he gave him a little correction and said, “You know, you’re going now.” And so the dog said, “OK,” — and they both stepped right into wet cement.
If the dog could have, lets say, pressed a button on his guide dog harness that would definitively say, “We need to go around this,” he could have saved him a potentially dangerous situation. If had slipped in the cement and fallen, or if it had been a hole or something like that that he could have fallen into, it would be really important for the dog to be able to tell him what’s going on, “We need to go around this, I’m not just being disobedient, we need to go around this.”
So other scenarios: hearing dogs. The way hearing dogs work is that they alert the owner to a sound, and they will run to the source of the sound. So if a baby’s crying, they’ll say put their paw on the owner’s knee and then run into the baby’s room and you’ll know that that’s the baby crying. Or if it’s a doorbell they will run to the door. But what if it’s a tornado siren? How does the dog tell you that that’s what that is? So you can train the dogs to differentiate the sounds, in fact we’re doing that right now with one of our demo dogs, and then they could press a different button the vest for different sounds so you’d know that it’s the tornado siren. So the dog could say, “That’s the tornado siren,” and the owner could make a much better decision on how to react.
Third scenario. Let’s say you have a search and rescue dog, right now the search and rescue dogs pretty much have to be controlled within line of sight because they have voice and hand signals, and things like that. And once the dog finds — let’s say they’re searching for a person — once the dog finds the person, they have a small padded stick hanging off of their collar called a bringsel. And what the dog does is it puts the bringsel in their mouth and then runs back to the handler to say, “Hey, I found it, I found somebody.” But what if that bringsel was electronic? And if the dog hit it, it would geolocate to a GPS satellite and tell the whole team, the rescue team, the handler, everybody exactly where that person was at that moment. And then the dog could stay with the person and do whatever needs to happen with that person until the team got there.
There’s all kinds of way that this could be used. And the more we think about it, the more uses we find for it.
BI: And how long have you been working on it?
MJ: It’s probably been about six or seven months that we’ve actually been engaged in the building the sensors and testing them. So a fairly new project.
BI: How do you train dogs to know what to do and where to move? What kind of sensors do you put on there? What kind of challenges do you find with that?
MJ: Great question. The good thing about assistance dogs and working dogs in general is that they’re usually already trained to do a lot of the things we need them to do, our sensors are based on the natural behaviour of dogs. So dogs can hold things in their mouth, they can bite down, they can take something and tug it, pull it, they can touch things with their nose. So all of the sensors that we’ve developed so far are just based on natural dog movement.
We have two different form factors for a bite sensor. So that bringsel I was telling you about with search and rescue dogs, we actually made an electronic brings-all, so that when the dog bites down on it, it actually sends a signal to a computer, a little microprocessor on the vest, and then we can do whatever we want to with that. We can hook it to a GPS, or we can have it emit a tone, or speak words, or whatever we need.
And then we have another form factor which is a little larger, sort of an oval shape, that’s a little easier for the dogs to grip. We also have a tug sensor, which is essentially a rubber ball at the end of a strap. The dog tugs on the ball with just enough force to activate the sensor. And then the last one is that we have already tested is a proximity sensor. So very similar to the paper towel dispensers that you might see in restrooms, where you wave your hand and the paper towel magically appears. We used very similar technology so that the dog waves their nose over it and it activates the sensor. And what was really fascinating about that is that the dogs, we started out teaching the dogs to touch the sensor with their nose, which is what I thought they’d be able to do, and the dogs all by themselves figured out very quickly that all they needed to do was wave their nose past it. So, we we’re very pleased with that, it was a “Wow!” moment.
BI: How did you figure out when the dog is truly understanding what you’re training it on and how much can you trust it? I imagine they get it right here and there, but they need to get it right 100% of the time, right?
MJ: You’re absolutely right. That is obviously one of the things that we’re going to be researching is what is the best way to ensure that the dog truly has an understanding of the task. And that’s really a dog-training problem. As a dog trainer, I’m quite confident that we can train the dogs to be 100% reliable on what they’re telling us. I really have no qualms about that.
BI: How far can this go? Right now, we’re talking about assistance dogs and a limited set of things. In five or 10 years, do you see this research expanding significantly to the point where anyone’s normal pet dog can be wearing some sort of sensor and getting trained and being able to say, “I don’t really like this food,” or “You have the music on too loud, and quite frankly it’s bothering me.” How much communication can this lead to between humans and dogs on a non-assistant, non-specialised level?
MJ: Of course we have enough ideas to easily fill 10 years of work. I love your examples, because those are indeed things that you might want to know.
Here’s another scenario: You’re at work, and your dog is home by himself, and he needs to go out, but he doesn’t have anyone home to let him out. He could press a button on his vest or on the door that would text you, “Hey, dog needs to go out.” You activate your home security system to open a doggie door, your video camera on the dog’s collar let’s you know that you’re dogs out there, if he’s done his business or come back in the house, and then you lock up your house again. I’d love to have that!
BI: That sounds great for anyone that has a pet.
MJ: Another thing, police have told us and home security is that one of the best things you can have in your home is a barking dog. What if that dog could press a button to alert you that someone’s in your yard? Now a dog might say “a squirrel’s in the yard” but almost all of my dogs, if there’s barking, there’s really somebody out there. So, if you could teach them to let you know, you turn on your video camera and say, “Yeah there’s somebody there, but it’s the mailman or that person isn’t supposed to be there, call 911.” That’s something you could use at home that I could see being a definite thing.
BI: Do you envision this as a business?
MJ: We’re in the early research stages, so we’re not looking to commercialize anything in the next year or two for sure. We just received funding from the National Science Foundation to do this work for the next two years, which we’re very excited about. So at the end of the two years we should have a much more viable and robust and accurate sensor.
I expect probably initially we’ll work with the assistance dog groups that are our partners on this project and also we’re discussing things with military training agencies for military working dogs.
This could be a huge game-changer for military applications, if you can imagine a bomb-sniffing dog that can tell you what kind of bomb it found. There’s also another side of this where the handler can send remote messages to the dog through electronics as well, such as maybe a small vibrating motor that’s inside the vest somewhere so that the handler could give a command like ,”Go sweep that side of the woods, go off to the right.” Sort of remote control for the dog. But also the handler might be able to give him a command by vibrating the middle of the chest meaning, “come back to me now.” So, the way military dogs are controlled right now is voice and hand signals, which means that they now have to be able to see their handler, which means their handler has to be out in the open makes them a sniper target.
So, being able to have electronics that the dog and the person can communicate with each other allows the handler to be safer and the dog to be safer. Right now, the way a bomb-sniffing dog finds a bomb, they lie down and they start barking. So the handler gets there and obviously that lets the enemy know that you found it and that you got a dog team on it. What we could do is the dog could find the bomb and geolocate, be able to press something on their vest or pull something on their vest that would connect with a satellite that would tell you where it was and then the dog returns to the handler. So the dog’s out of danger, the handler is never in danger, they can stay hidden the whole time. So we’re thinking this can be a huge safety procedure that would enhance the quality of what we can do on the ground with dog teams and keep them out of harm’s way as much as possible.
BI: Why now? Is there new technology that enables something like this right now?
MJ: We’ve finally gotten it small enough and powerful and with good enough battery life that it’s feasible to really use this stuff, get it out of laboratory, put it on and wear it.
BI: How would a human communicate with a dog? Would it be through their smartphone? Through Google Glass?
MJ: The guide dog might not need the handler to wear anything if they don’t need to communicate to the dog anything through the FIDO system or through the vest. You might just have a speaker on the vest that would say, “We need to go around this,” or something that would go to a Bluetooth earpiece, which is what we’re going to be implementing, but you wouldn’t necessarily need something as powerful as the Google Glass for something like that.
However, you can certainly see a hearing dog, that would make a lot of sense that you could have a text message that would come up in your head-mounted display that would say, “Hey, I just heard the tornado siren, we better do something about this.” It sort of depends on the application.
BI: How else can you get the dogs to communicate with humans?
MJ: There are a couple of other things that we’re looking at for these sensors. One thing is, if we can put sensors on the dog that senses body position of the dog, there are two things we can do with that. One is we might be able to interpret natural dog language, which is largely postural.
Dogs communicate very differently than we do. They don’t necessarily talk like we do, but if they bow down with their ears up and their tail up and wagging, that’s called a play bow and that means they’re happy, but if they stand up on their tip-toes and and their tail is still that’s an aggressive posture.
We might be able to interpret natural dog language with these sensors, put these on and tell what the dogs are doing.
Another thing we could do is teach the dogs to do specific gestures, like that bow is something my dog knows how to do, that might be instead of pushing or activating a sensor, just having the dog go into a position that it might not normally go into to communicate something. It’s called activity recognition and that’s another way that we’re going with those.
The third thing that we can do with that is possibly temperament testing to evaluate dogs that might be suitable for different types of work. So you need a dog that’s got some aggression for police dog work, you can see if the dog has tendency towards that with the sensors. If you need one that’s very docile and friendly for assistance dog work, you might be able to evaluate the dog’s temperament. These are just ideas that we’re throwing around, and with the technology that we’re developing should be something that’s possible.
BI: It’s interesting to think about how posture is so important.
MJ: In many, many years of dog training, my experience is that a lot of people anthropomorphize their dogs.
A classic example: you come in, and somebody has messed on the floor and you find the dog and the dog looks cowed and the dog’s sort of crawling through the room looking cowed and so people say, “He’s guilty, he knows he did that and he’s guilty and he knows he should be punished.” Well that’s not true at all. What the dog is reacting to is your anger. The dog doesn’t remember he messed on the floor, he doesn’t even know he did it, he doesn’t remember that, all he knows is that you came in and you’re mad so what he’s trying to do is calm you down and that’s a calming signal.
A dog being submissive, putting his head down, putting his tail between his legs, that’s a submissive posture, and what he’s trying to do is calm you down. He doesn’t know why you’re upset. So that’s the main thing, that a lot of people misread dogs’ language and don’t really know what that’s all about. So it might lead to a better understanding of what our dogs are really saying to us, “I’m frightened of what you’re saying right now, of the way you’re yelling right now. I’m not remorseful because I ate the bacon on the counter.” They don’t even remember that they did that.
So I think people think they know their dogs, but a lot of people do misinterpret it.