Some days Google seems like it’s more of a science fiction factory than a search engine, developing products like driverless cars, and augmented reality glasses. An academic project at Berkeley adds another element to the mix – Robots. Robots that can help pick up commonplace objects around your home, and put them in their proper places.
A paper submitted to the IEEE International Conference on Robotics and Automation, to be held in Karlsruhe, Germany on May, 2013, describes the role that Googles visual search queries plays in helping robots understand the objects that they might try to pick up, before they do. In Cloud-Based Robot Grasping with the Google Object Recognition Engine, we’re told about cloud-based robots that can view objects, and send queries about them to version of Google Goggles on the cloud to learn more about those objects and the best way to grasp them.
Google Goggle’s is Google’s visual search app, which enables you to take photographs and send them to Google to potentially perform facial recognition searches, OCR searches for text in images, product and bar code recognition, recognizing landmarks and places and named entities, and more. I spent a few hours at my Mom and Dad’s house a couple of weekends ago taking pictures of almost every photo and painting they had on their walls, and seeing if Google Goggles recognized any of them.
Another feature that the visual search engine is capable of is recognizing objects, and the Berkeley team, with the assistance of James Kuffner of Google, appears to have achieved a goal that had eluded them in the past with the use of Google Goggles. From the paper’s introduction:
One as-yet unachieved goal of robotics and automation is an inexpensive robot that can reliably declutter floors, tables, and desks by identifying objects, grasping them, and moving them to appropriate destinations such as shelves, cabinets, closets, or trashcans.
Errors in object recognition can be costly: an old chocolate bar could be mistaken for a cellphone and moved to the charging station, or vice versa â€” a cellphone could be placed in the trashcan. The set of objects that may be encountered in an unstructured environment such as a home or office is essentially unbounded and dynamically grows as our increasingly capitalist global economy designs new products to satisfy demand from consumers (and shareholders).
The project used a customized version of Google Googles to perform object recognition, training the system on 15 commonplace objects that tend to be around a home, such as a bar of mustard, an air freshener, and a bar of soap.
When I was writing about the future of search being spoken and visual queries not long ago, I hadn’t anticipated that some of those searchers might not even be human.