Craig Doty II, a Tesla owner, narrowly avoided a collision after his vehicle, in Full Self-Driving (FSD) mode, allegedly steered towards an oncoming train.
Nighttime dashcam footage from earlier this month in Ohio captured the harrowing scene: Doty’s Tesla rapidly approaching a train with no apparent deceleration. He insisted his Tesla was in Full Self-Driving mode when it barreled towards the train crossing without slowing down.
Did you watch the video? It was insanely foggy there. It makes no difference how big the obstacle is if you can’t even see 50 meters ahead of you.
Also, the car did see the train. It just clearly didn’t understand what it was and how to react to it. That’s why the car has a driver who does. I’m sure this exact edge case will be added to the training data so that this doesn’t happen again. Stuff like this takes ages to iron out. FSD is not a finished product. It’s under development and receives constant updates and keeps improving. That’s why it’s classified as level 2 and not level 5.
Yes. It’s unreasonable to expect brand new technology to be able to deal with every possible scenario that a car can encounter on traffic. Just because the concept of train in a fog makes sense to you as a human doesn’t mean it’s obvious to the AI.
To me it seems you just spent three paragraphs answering your own question.
If I couldn’t trust a system not to drive into a train, I don’t feel like I would trust it to do even the most common tasks. I would drive the car like a fully attentive human and not delude myself into thinking the car is driving me with “FSD.”
You can’t see 50 meters ahead in that fog.
Completely true. And I would dictate my driving characteristics based on that fact.
I would drive at a speed and in a manner that would allow me to not almost crash into things. But especially trains.
I agree. In fact I’m surprised the vehicle even lets you enable FSD in that kind of poor visibility and based on the video it seemed to be going quite fast aswell.
LIDAR can
Yeah there’s a wide range of ways to map the surroundings. Road infrastructure, however is designed for vision so I don’t see why just cameras wouldn’t be sufficient. The issue here is not that it’s didn’t see the train - it’s on video, after all - but that it didn’t know how to react to it.