Cambridge Engineers Are Teaching Your Smartphone to Think Like an Autonomous Car

Jennifer van der Kleut

A team of researchers from Cambridge University are training smartphones to “see” the world like an autonomous car would-even when it can’t pick up a GPS signal.

Reports Gizmodo, the team has developed two different software programs that can run on mobile phones and teach the phone to think like a self-driving car.

The first, SegNet, takes the street scene in front of you and classifies all objects it sees, into 12 different categories: Sky, Building, Pole, Road Marking, Road, Pavement, Tree, Sign Symbol, Fence, Vehicle, Cyclist and Pedestrian.

According to Gizmodo, testing is currently at 90-percent accuracy, and even works with light, shadow and night-time environments.

The second program uses elements of the street scene to discern the phone’s location. Much like the system used by Google’s self-driving cars, the program uses the geometry of buildings and other factors and compares it to existing maps to determine the location.

Current testing is going so well, the team says it is operating in the Cambridge area “far better than most GPS systems.”

While the programs will likely not be able to work with cars to turn them into autonomous vehicles, the team says it could theoretically work with “domestic robots.”

Read Gizmodo’s full article and see video of how SegNet “sees” the environment.