Object detection and classification is step zero in autonomous driving, but accurately predicting behavior and intentions of entities in the scene involves heavy machine learning and a hybrid approach including a combination of different types of ML models, model-free approaches, and physics-based systems.