AI Trained to Act Like a Dog

Not only are dogs man’s best friend, they also provide all sorts of services from guiding the visually impaired, locating lost people, and sniffing for drugs. Artificial intelligence may not immediately be able to perform these tasks, but a group of researchers are training AI systems to be more capable.

Kiana Ehsani from the University of Washington in Seattle and her colleagues built a database of dog behaviors to train AI to make dog-like decisions. They investigated how AI could act and plan like a dog and what information they could learn from canines.

Researchers attached measuring sensors to a dog to record how it behaved—and then trained an AI system to act like a dog.

To do this, they equipped a single dog with inertial measuring units on its legs, tail, and body that recorded the relative angle and absolute position of each body part. The researchers also attached a GoPro camera on the dog’s head to record its surroundings and ambient sounds. In all, they collected a total of 24,500 video frames with matching body position and movement data. They used 21,000 of these frames to train the AI system and the rest for validation and testing.

Their goal was to build an AI that could predict a dog’s movements given a sequence of images. The AI did this by looking at video frames and studying what the dog did next. It also had to learn from the dog’s behavior—such as recognize what where a dog could and couldn’t walk. Here’s MIT Technology Review:

That’s interesting work that shows how AI systems can match certain types of animal performance. “Our model learns from ego-centric video and movement information to act and plan like a dog would in the same situation,” they say.

And dogs are just the beginning. There’s a planet full of animals to study, each with their own unique intelligences and ways of approaching the world. Someday, AI may even help us understand them better.