Robots In The Cloud

With the phrase "web 2.0" falling out of vogue, the most exciting new uses of the internet are now all about the cloud, a term for servers invisibly doing smart, fast things for net users who may be on the other side of the world.

But it's not just humans that stand to gain, as a recent corporate acquisition by cloud pioneer Google demonstrates. Google has snapped up British start-up Plink, which has devised a cellphone app that can identify virtually any work of art from a photograph. Plink's app will bolster Google's Goggles service, which uses a cellphone camera to recognise objects or eventranslate text. Unlike most cloud start-ups, Plink sprang from a robotics lab, not a Californian garage. Its story demonstrates how the cloud has as much to offer confused robots as it does humans looking for smarter web apps. 

Spatial memory

Mark Cummins and James Philbin of Plink developed the tech while working in Paul Newman's mobile robotics research group and Andrew Zisserman's visual geometry group, both at the University of Oxford. The group is trying to enable robots to explore the cluttered human world alone. Although GPS is enough to understand a city's street layout, free-roaming robots will need to negotiate the little-mapped ins and outs of buildings, street furniture and more.

Image-recognition software developed at Oxford has helped their wheeled robots build their own visual maps of the city using cameras, developing a human-like ability to recognise when they have seen something before, even if it's viewed from a different angle or if other nearby objects have moved.

You are here

Plink gives cellphone users access to those algorithms. Photos they take of an artwork are matched against images on a database stored in the cloud, even if they were snapped from a different angle. Although the Oxford team's algorithms originally ran entirely on the robot, Newman is now working on moving the visual maps made by a robot into the cloud, to create a Plink-like service to help other robots navigate, he says. Like a user of Plink, a lost robot would take a photo of its location and send it via the internet to an image-matching server; after matching the photo with its map-linked image bank, the server would tell the robot of any matches that reveal where it is.

Newman is also testing that concept using cameras installed in cars. "We can drive around Oxford at up to 50 miles per hour doing place recognition on the road," he says.

If image maps from many cities were made into a cloud-like service, any camera-equipped car could look at buildings and other roadside features to tell where it was, and the results would be more accurate than is possible with GPS.

Adept users

Adept Technologies of Pleasanton, California, the largest US-based manufacturer of industrial robots, is also looking cloud-ward. Some of the firm's robots move and package products in warehouses. With access to a Plink-like image-recognition system they could handle objects never encountered before without reprogramming.

"This connection of automation to vast amounts of information will also be important for robots tasked with assisting people beyond the factory walls," says Rush LaSelle, the company's director of global sales. A "carebot" working in a less controlled environment such as a hospital or a disabled person's home, for instance, would have to be able to cope with novel objects and situations.

Cellphones, humans and robots all have a lot to gain from a smarter, faster cloud.