A.I. System Could Train Your Dog While You’re Away From Home
One of the few good things about locking and working from home was spending more time with pets. But when the world goes back to normal, people go back to the office, and in some cases it means dogs stay at home for a large part of the day, hopefully with someone coming into your house to let them out at noon period.
What if an AI device like a next-generation Amazon Echo could give your pooch a dog training class while you were away? That is the basis for a project carried out by researchers at Colorado State University. Originally discovered by Chris Stokel-Walker, author of YouTubers: How YouTube Shook TV and Created a New Generation of Stars, discovered and reported by New Scientist, the work is a prototype that can issue dog commands to verify whether they are present are obeyed and then give a reward as a reward, if they are.
“We developed a device that uses machine learning to monitor and reward positive dog behavior,” Tom Cavey, a co-author of the project, told Digital Trends. “If a dog shows a desired action for a long time, the device releases a reward. It does this in real time using a tiny embedded, low-power computer called the Nvidia Jetson Nano. With the device we can record live video and use it in a machine learning model to find out what the dog is doing in real time, completely independent of local or remote connections. If the dog displays certain behaviors such as “sitting” or “lying down”, a good behavior is noted and a treat is dropped from the device. “
AI can do the picture classification better and better. In this case, however, using a disconnected embedded device meant handling memory and hardware constraints that hampered researchers’ ability to use complex machine learning models. They used optimization and quantization techniques to, among other things, reduce the model size so that it could run on the Jetson Nano.
Jason Stock, co-author of the project, told Digital Trends if they have any plans to commercialize the technology: “Our approach consists of inexpensive hardware and software components that would enable a relatively affordable product to be commercialized. We are considering the possibility of evolving this in the face of a larger budget to create a better-functioning prototype and improving the model and dataset to identify more behaviors – maybe even those that are perceived as negative. “
In other words, there is a chance that this will be for sale at some point. But first you have to sit, stay, and wait a while.
An article describing the work was recently published on arXiv, the open access repository for electronic preprints.