Hacker News new | ask | show | jobs
by jqpabc123 522 days ago
The real elephant in the room is whether FSD is even possible by using primarily visual input.

Personally, I don't think so.

There are simply too many driving situations where visual input is severely limited. FSD can use all the help it can get.
2 comments
jfengel 520 days ago
Don't you drive using primarily visual input?

I don't know if machine learning can ever match the human brain for that. The brain does a lot of fairly advanced inferences that require a deep understanding of the world and the people and things in it.

Still, I'm not sure how much additional inputs would help the ML. If you had to drive by "touch" (LIDAR), you probably shouldn't be allowed to drive. It might be useful when the visual system has failed, to stop the vehicle before it hits something, but if the visual system fails that often then the system wouldn't be usable for any purpose.

link
jqpabc123 520 days ago
Don't you drive using primarily visual input?

Visual input combined with a lot of mental projection and inference and understanding/experience and good split second decision making.

Examples: How does a FSD vehicle use it's camera to identify "black ice" on an overpass? How does it identify that a heavy truck tire has just exploded in it's lane a few cars ahead on the freeway? Or how about a bumper dropped from a car ahead or a large truck losing it's load?

All these and more have happened to me and I lived to tell about it.

link
senordevnyc 519 days ago
Still, I'm not sure how much additional inputs would help the ML.

This is a wild take when Waymo is doing 150k+ rides per week with no one in the driver seat.

link
LittleTimothy 521 days ago
Does that matter? Waymo solved FSD. The technical detail of the implementation is moot.
link
Zigurd 521 days ago
Waymo's approach to autonomy, and robotaxi service, are both very different from Tesla's concept of a plan for general purpose autonomous driving plus an AirBnB-like taxi fleet.

In addition to what will turn out to be a foolish fixation on cameras as the only sensor, Tesla FSD hasn't got nearly as much mapping data, or real-time traffic data which Waymo can source from Google Maps, or in-vehicle monitoring of passengers and vehicle condition.

Elon has his online claque for distorting reality. But reality is still there. That's why Elon wants to buy the whole government to get bigger reality-denial tools.

link
jqpabc123 521 days ago
The technical detail of the implementation is moot.

Waymo solved FSD using LIDAR and other sensors that Tesla has publically rejected in favor of visual cameras.

The technical detail is not at all "moot" due to the obvious fact that Tesla has yet to achieve what Waymo has.

https://www.forbes.com/sites/bradtempleton/2022/10/31/former...

link