Tesla recently revealed on X that its Full Self-Driving (FSD) system can now anticipate pedestrians' intent to cross the road—rather than waiting until they actually step onto the street before initiating braking. This capability stems from the "end-to-end" neural network architecture introduced with FSD V12 and further enhanced in versions V13 and V14 with Transformer-based temporal processing techniques, creating a roughly 15-second "time buffer." This allows the system to infer pedestrian behavior even when the target is temporarily occluded, by analyzing historical trajectory data. Additionally, FSD version 14.2.1 can recognize unstructured social cues such as traffic police hand signals, leveraging an upgraded "skeleton tracking" model that captures movements of 18 human body joints. Despite these technical breakthroughs, FSD still faces regulatory scrutiny and localization challenges, particularly requiring further optimization for China's complex driving conditions.