Why has Tesla merged perception and behavior?

Waymo's approach, which has been publicly shared multiple times even as the algorithms change, is a more traditional approach to robotic systems. Sensor data comes in, and a perception system is responsible for understanding the world state. This includes fusing sensor data, filtering necessary information, understanding objects, segmenting, tracking, and getting states of things like traffic lights, lane markers, vehicles, and their lights. It may also involve semantics, such as where a person is looking or their body orientation, and understanding construction zones in unstructured environments. Anything that helps make better decisions is part of perception, which is about seeing and understanding the world around you. Then you have the behavioral side, traditionally split between behavior prediction and planning. Behavior prediction involves predicting what pedestrians, cyclists, cars, buses, and others will do, which is crucial for deciding your actions.