Google reveals how its self-driving cars distinguish drivers from cyclists

Google has revealed in its latest report how its self-driving cars can tell drivers apart from cyclists and other users of the road.

Like in many of its products, machine learning algorithms figure prominently in the car's detection technology. After "seeing" many examples of bicycles from every angle, with its cameras and sensors, the car's computer had been able to learn how bicycles look like.

"Our software learns from the thousands of variations it has seen - from multicolored frames, big wheels, bikes with car seats, tandem bikes, conference bikes, and unicycles," Google said in its report, published yesterday.

Machines have to be taught that a bicycle was an object that was distinct from the rest of the environment, not always an easy task and even when Google's driverless car could ''see'' a cyclist, it did not always know what  with that information.

Situations in which drivers and cyclists mingled, presented challenges for autonomous cars as all the ordinary rules of the road seemed to fly out of window.

The report revealed that if the car registered both a bicyclist and a parallel-parked car with an open door, it would move over to give the biker some space.

The system was also capable of detecting hand signals, and it could remember a specific cyclist's hand signals, in case a person was signalling 500 feet ahead of a turn, while another may signal only 100 feet before. It could also manage its way through a unique situation, when one cyclist veered in front of Google's car while another turned and started riding against the flow of traffic.

The report also includes collision updates, and in June it recorded two, with the first involving another car veering from its right-turn lane and contacting a self-driving Google car in an adjacent right-turn lane. The second involved another car moving forward and tapping Google car's rear bumper. Both accidents occurred in Austin, Texas.