Waymo Gives Away Free Self-Driving Training Data -- But With Restrictions

Yesterday, Waymo announced it would "open" a large dataset of self-driving training data. This gathered attention because Waymo has, by a huge margin, the largest number of self-driving miles under its belt, and thus one of the most envied collections of tagged data that can be used to train and test neural networks, one of the key tools used in building robots and self-driving cars. People setting up to build a self-driving car almost universally use machine learning techniques. With machine learning for computer vision, you provide the computer with images that a human being has already put labels on, saying what in the image is a car, or pedestrian, or road surface. Give the computer enough, and your machine learning technique -- today, most commonly a convolutional neural network -- will use advanced statistical techniques to come to a more general understanding of what distinguishes the various components.