FSD didn't handle this situation well. The driver had to step on the accelerator to get past a car blocking the lane, and FSD didn’t return to the correct lane afterward. I added English subtitles for key conversations.

Source: Douyin 影音时光

What REAL customers say about TurboScribe's unlimited AI transcription 👇 🇺🇸 "Thank you for developing such a fantastic AI tool." - Kathleen 🇮🇹 "The AI transcription tool I've been waiting for." -Luca 🇷🇴 "It's so accurate that it's scary!" - Cosmin

@alojoh These edge cases are good. Hopefully the feedback makes it way back to Tesla to fix.

That is the case Tesla wants to collect.

It noticed that there are crazy things happening on the right lane so it decided to use left

Not saying it handled that correctly, but if you look closely this LOOKS like a vehicle with flashing lights blocking a lane. And you would usually see that at construction in which case FSD did the completely correct thing to not use that lane, as it would have workers on it. This is actually a very interesting edge case that shows why non-neural network approaches will never succeed. The real world is weird.

@aelluswamy

Why the left lane is wrong lane if it’s one way road. Maybe not the preferred lane for the driver but not that wrong. If another car come on same lane behind then the FSD will go to right lane.

Good Videos thanks for the effort

@alojoh Nice new data 🤗 @aelluswamy @Tesla_AI

They printed should have disengaged, otherwise it is harder for the model to learn