Everything from sequencing genomes to writing sick rap beats can be tackled by deep learning software nowadays. By feeding in databases of information, we're helping AIs learn better over time, whether that means helping autonomous cars process their environment, tracking hands in virtual spaces, or letting barista-bots figure out how to use a new coffee machine on their own. Apparently, it can also help keep cats from pooping on your lawn.

This high-tech solution to an everyday problem comes from the mind of a Nvidia systems software engineer by the name of Robert Bond. He's the kind of guy who builds an automated laser-firing "Ant Annoyer" to entertain his grandkids. Motivated by the dual desires to learn about neural net software and keep his lawn cat poop-free, Bond rigged up a system that detects when cats – and only cats – wander into his yard, and responds by turning on the sprinklers.

The cat-chaser is quite clever. A Foscam IP camera keeps watch over the yard, and when it detects motion it takes a photo once every second for seven seconds. These are sent to an Nvidia Jetson TX1, a development module designed to run this kind of prototypical hardware system. The Jetson is running a "fully-convolutional neural network for semantic segmentation", or FCN, which has been fed as many images of cats as Bond could get his hands on – which if the internet's love of cats is anything to go by is sure to be a lot.

The network scans the images for signs of what it's been taught cats look like, and if it detects one, the Jetson pings a Wi-Fi card connected to the home's sprinkler system. On goes the water, off go the cats. The deep learning system makes sure that only cats are sprayed, so visitors to the house won't be in for a sudden shower – with the possible exception of trick or treaters dressed up as Garfield or Catwoman on halloween.

"It wasn't actually that much work," Bond says of the system, which he put together in around 10 to 15 hours. "The new Jetson TX1 is really good at running these neural nets."

And it's working: the AI apparently isn't the only thing learning, as Bond says it only took a few days for the neighbourhood cats to decide to find a new hangout. He outlines the system on his website, and has put the code and firmware up on GitHub for any curious electronics enthusiasts who may have their own pest problems.

With that up and running, Robert Bond is working on his next project. Since the system can determine where in the yard the cats are, Bond wants to send targeted remote-controlled cars out to spook the intruders when the cat detector is triggered. Deep learning just has so many applications.

Source: Nvidia blog

View gallery - 3 images