A machine designed to find a children’s book character is causing a stir on social media. “There’s Waldo” is a robot that uses computer vision to locate the beanie-clad chap in the “Where’s Waldo” series of books, automating one of the great stresses of five-year-olds worldwide.
The machine was created by the creative agency Red Pepper, which demonstrated its creation in August 2018. It uses Google’s AutoML Vision, a cloud-based artificial intelligence service that enables developers to create custom models for identifying images. The team uploaded 26 drawings of Waldo to train the model before setting up the bot. The machine uses a Raspberry Pi computer equipped with the PYUARM Python library to control the UARM Metal. The arm uses a Logitech webcam and OpenCV to identify faces and send back to Google, placing the rubber hand down if Google reports a 95 percent or higher confidence rate. The robot was shared by Twitter user “CKYPT” last week, and it has since received nearly one million views:
The results are impressive. Its highest record for finding and identifying a match is 4.45 seconds, much faster than it normally takes a kid to complete the task. Ditching the robot from the equation could make the process even faster: a system outlined by Machine Learning Mastery in 2014 described how developers could use OpenCV, Python and Template Matching to identify Waldos in less than a second.
It may seem like a convoluted way of taking the fun out of a kids book, but the underlying technology has a number of important uses. Computer vision is helping to power autonomous car projects like Cognitive Pilot, where the system identifies obstacles and takes appropriate action. An MIT research project trained an A.I. to recognize food recipes based on sight, after training on one million recipes and 800,000 pictures of food. Boston Dynamics’ SpotMini is also using computer vision to train robot dogs so they can traverse a house, pick up a can and put dishes in the dishwasher, all of which is a touch more useful than finding Waldo.
AutoML could offer a glimpse at how future A.I. systems will operate. Google Brain researcher Quoc Le told Synced that AutoML’s straightforward interface means “we’re entering the fourth generation of machine learning: the machine learning so that you don’t have to do much, it will learn everything.”
That learning could even extend to the location of Waldo.