XPeng has begun gradually implementing end-to-end solutions 🌹. The pre-end-to-end autonomous driving solutions were 'rule-based,' much like a driving instructor teaching parallel parking: stay 30cm from the line, reverse until the rear wheel touches the line, then turn the steering wheel all the way to the right, and straighten when a certain wheel touches the line... In contrast, end-to-end is like an experienced driver who has seen all kinds of parking spaces and methods, long forgotten the mnemonics, and relies on neural network 'intuition' to park with a single press of the accelerator.

In the same scenario, the difference before and after end-to-end implementation is striking. As @XP-何小鹏 mentioned in the review video, the old U-turn scenarios felt like rules switching one after another, while end-to-end enables smooth, human-like U-turns.