Part 4/12:
While it's tempting to think of FSD as a single monolithic neural network, Tesla actually separates it into perception and planning components.
Perception analyzes sensor data to detect objects, road signs, markings, pedestrians, and other critical elements.
Planning takes this information along with navigation goals to determine immediate actions, such as maneuvering through intersections, adjusting speed for weather conditions, or parking.
This separation allows Tesla to develop specialized sub-models or experts trained for specific scenarios—such as highway driving versus city navigation or driving in snow versus clear weather. These experts are integrated within the overall neural network system, activated as needed based on the current environment.