Google I/O 2018 has shed some light on how Waymo is using AI to make fully self-driving cars a reality. John Krafcik, CEO of Waymo, revealed the new milestones the company has achieved in the field of autonomous car technology — indicating people can soon book a driverless car in the near future.
The engineers at Waymo have teamed up with Google Brain team to apply Deep Nets to the pedestrian detection system which has reduced errors for its self-driving cars. The pedestrian recognition issues have been reduced by 100 times, making the system safer and more capable on the road.
Waymo uses TensorFlow ecosystem and Google’s data centers to train its neural networks. The ML models are tested in simulation, leading to 15 times more efficient learning and rapid deployment of the latest nets on the self-driving cars.
Other factors like snow or rainfall that interfered the car’s sensors are not a problem anymore for the self-navigating car as machine learning helps in filtering out the noise and correctly identify pedestrians and vehicles on road.
It can easily navigate through difficult situations like maneuvering through construction zones and analyzing abnormal behaviors of other cars to avoid a collision. So, in some ways, Waymo could be safer than cars that have a driver.
Read our complete Google I/O coverage.