Probabilistic neural network for predicting hidden context of traffic entities for autonomous vehicles

This invention describes an autonomous vehicle that uses a special kind of artificial intelligence, called a probabilistic neural network, to guess what other cars and people in traffic are likely to do next. It takes a picture of the road, figures out the likely behaviors of traffic entities, and importantly, also calculates how sure it is about those guesses. The vehicle then uses these predictions and their uncertainties to navigate safely. The claims specifically detail that this AI is trained using a method that involves comparing its predictions to human observations and calculating a specific statistical measure called an 'evidence lower bound'.