IntentNet: Learning to Predict Intention from Raw Sensor Data

Autonomous driving is one of the most exciting problems of modern artificial intelligence. Selfdriving vehicles have the potential to revolutionize the way people and freight move. While a plethora of systems have been built in the past few decades, many challenges still remain. One of the fundamental difficulties is that self driving vehicles have to share the roads with human drivers, which can perform maneuvers that are difficult to predict. Human drivers understand intention by exploiting the actors' past motion as well as prior knowledge about the scene (e.g. the location of lanes, direction of driving).