Image via Twitter: @ecelectronicsuk
AI-powered cat door flap prevents pets from entering with dead prey
An Amazon employee created an AI-powered pet door that prevents his killer cat from bringing dead critters into the house.
Image via Twitter: @ecelectronicsuk
Technology is all about making our lives easier, helping us deal with the world in new and better ways, and helping to deal with the pain points in our lives.
For one Amazon employee, his pain point was grey, white, fluffy and partial to the occasional murder.
Gory little presents
Ben Hamm had been the recipient of one dead animal carcass too many when he decided that his cat Metric needed to learn a lesson about appropriate gifting.
If you have no idea what I’m talking about right now, you’re not a cat person. It’s okay, I forgive you.
Cats so you know have a tendency to go out into the world, murder and mutilate birds and rodents; and then bring their tiny tortured corpses home as a present for their human parents.
A popular gifting area is inside shoes or slippers. Yes, cats are fluffy little psychopaths, but at least they’re fluffy.
Hamm, however, decided that he’d had enough of Metric trying to add to its dead animal collection.
The purr-fect cat flap door
He built a DIY AI-powered cat flap that would lock Metric out for 15 minutes if the cat was approaching the flap with prey in its mouth.
Hamm used Amazon AI tools DeepLens and Sagemaker to power his DIY project.
DeepLens is a camera that is specifically engineered for machine-learning applications. Sagemaker is a service for users to buy or build machine learning algorithms and then train them.
It’s easier to train AI than to train a cat
According to Hamm, the most laborious part of the project was training the model. He supplied over 23 000 photos.
He needed to show images with the cat in view, coming toward the camera and moving away from the camera. As well as with and without prey in his mouth to train the model.
The project used supervised learning. Essentially, the model is fed a large amount of labelled information and taught to recognize patterns and act accordingly.
Once it has enough examples, it can apply its ‘experience’ to new situations. Over five weeks with the program running, Hamm claims that Metric was only locked out incorrectly once. That is pretty impressive.
A software engineer did suggest that it might have been easier to train the cat than to train the machine learning model. But we can only assume that the software engineer never tried training a cat to do anything.
