Humans and dogs make good teams. Working dogs have helped humans in one way or another for something like 15,000 years, and fortunately for us, dogs have been consistently clever enough to translate human modes of communication into commands that they can understand. They're able to respond to auditory signals (including spoken language and things like whistles) as well as visual signals like hand movements.
This is all well and good if you're in close proximity to your dog, and if you're able to take advantage of those modes of communication yourself. But for search and rescue dogs and military working dogs that may operate beyond line of sight, or for people with some kinds of disabilities, neither of those communication options might be available.
In Amir Shapiro's lab at Ben-Gurion University in Israel, researchers have been training dogs to respond to haptic cues from a vibrating vest. Or rather, they've been training one specific dog, named Tai, who has been a very good boy about the whole thing and has successfully shown that wireless remote control of your pupper is both possible and practical.
Tai (a six-year-old lab and German Shepherd cross) is wearing a vest with embedded haptic vibration units. The general principle is the same as having your phone on vibrate in your pocket—you can feel a buzz when it goes off. Tai's vest has four units in it: front and back units, and left and right units, all controlled wirelessly. Using positive reinforcement in the form of treats, Tai was verbally taught to associate four commands with distinct 1.5-second-long vibration patterns: a constant vibration in the front right means spin, a pulsing vibration in the front right means walk backward, pulsing in the front left means approach, and constant vibration in both rear units means lie down.
It took less than an hour to train Tai for each task, and he did just as well following the haptic commands as he did following verbal commands (better, in some cases). The researchers suggest that dogs could be taught to follow a wide variety of commands using haptic vests like these, with potential benefits in several scenarios:
Non-vocal communication may prove beneficial in many cases, such as discrete contact with [military working dogs], increasing capabilities of [search-and-rescue dogs] and other working dogs, reconnecting with run-away pets, communication by speech-impaired handlers, and even communicating with deaf dogs. Our current proof-of-concept study shows promising results that open the way towards the use of haptics for human-canine communication.
For more, we spoke with first author Yoav Golan via email.
IEEE Spectrum: Can you tell us more about Tai?
Photo: Dror Einav
Yoav Golan:I raised Tai as a seeing eye pup, and got him back after he failed the seeing eye course. He is easily distracted by dogs and cats, has a tendency to pull with force on walks, and has a "sniffing problem" (he likes to sniff a lot on walks). Tai is one of nine in a litter bred by the Israel Guide Dog Center for the Blind. A fun fact is that only one of the nine siblings became a seeing eye dog, and that dog, named Tango, was raised by my co-author Ben Serota. That is actually how we knew each other, and what drew us to do the project together.
What was your experience like training Tai to respond to haptic commands?
It was surprisingly easy. I am not a professional trainer in any way, but it still did not take long to teach him the different commands. It is important to note that Tai already knew the four commands vocally, meaning that the training was more of a "translation" to haptic commands, rather than training a new command altogether. Still, I don't see any reason that haptic commands would be any more difficult than vocal commands, and possibly might be easier. Vocal commands are difficult because they are inherently inconsistent (we never say things twice in the exact same way), and are mixed with a lot of noise (other words, and actual noise). Haptic commands are much clearer in that respect, and therefore may be easier to train, but we haven't tested this yet.
Are the number of different haptic commands and the complexity of those commands similar to what one can achieve through verbal communication with a dog?
While we haven't tested the upper limit of haptic training cues, in my opinion a dog can learn as many different commands as it can distinguish the difference between. Meaning, I don't see any reason that if a dog can learn 100 vocal commands, it can't learn 100 haptic commands, providing it can tell them apart. If we assume that a dog can understand more modalities than tested (i.e., tell the difference between further temporal modulations, or mixed spatial and temporal commands), this number rises hugely. Furthermore, we used four motors out of convenience, and there's no reason I see [to not] add more motors at other locations on the dog's body (around the neck, along the spine, at the belly), which adds even more possibilities.
Can you elaborate on what "fully or partially autonomic dog training" means?
This is kind of science-fiction-y, but there are dog vests that recognize a dog's position using accelerometers (lying down, sitting, running), and can dispense treats to the dog. You could integrate the two vests, and a computerized training regime, to get autonomic training. Meaning, you could put a vest on a dog, set it to train the dog "sit," and the vest would detect the dog's pose, issue a command, and reward the dog if it sits. Eventually the dog will learn to associate the command with sitting, without human intervention.
Dogs in pounds are more likely to be adopted if they are basically trained, so you could conceivably use a vest like this for a lot of dogs in a pound at once, without having to hire expert trainers, which are expensive and hard to find (which is why very few pounds try to train their dogs). Since the commands are haptic, two dogs can be trained nearby each other without interfering in each other's session. A human can be involved to supervise, or to train a dog more efficiently. For instance, if you want to train a dog to follow commands without a line of sight, you could issue the command remotely, see if the dog followed the command (via the sensors), and give the dog a reward—all from a distance and without a line of sight.
A Vibrotactile Vest for Remote Human-Dog Communication, by Yoav Golan, Ben Serota, Amir Shapiro, Oren Shriki, and Ilana Nisky from Ben-Gurion University of the Negev in Israel, was presented at the World Haptics Conference on 12 July in Tokyo, Japan.
Evan Ackerman is a senior editor at IEEE Spectrum. Since 2007, he has written over 6,000 articles on robotics and technology. He has a degree in Martian geology and is excellent at playing bagpipes.