Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Ross Gerber shows Dan O'Dowd that FSD is much safer than a human

This site may earn commission on affiliate links.
too many software engineers in this thread sidetracked on whether FSD would have stopped or not.

the bottom line is that with only feet left between the car and the stop line, regardless if a human brakes or the car brakes, that is going to be very forceful deceleration, way more jarring than any phantom braking event. Let's say the car stopped in time. Are we really going to pat FSDb on the back for doing a good job? Would your passengers be marveling at the technical prowess of the self-driving system? And extrapolate this to a future robotaxi application. As you're napping, or working on your laptop, and suddenly there's a 1g deceleration event for a situation that could have been executed smoothly. How would you perceive that self-driving system?
First of all, safety trumps comfort. If the system is shitty, it may need to apply max brakes to minimize damage.

Secondly, this is not an occluded sign. There are large STOP letters painted on the road and there is a warning sign of the upcoming stop sign 10 secs before the stop sign.
We need to be looking at this from a user experience standpoint.
I completely disagree. Tesla cannot drive reliably yet. Let's talk UX when Tesla has 1/5 of Waymos reliability at 8k miles per DE. Right now that would be a 500x improvement, which will likely never happen at the current rate of progress.
At the same time, we should be careful about criticizing vision only. When you have a blind stop sign (I've seen way worse than the Ross example), the addition of lidar or radar would not help at all - both would be foiled by the trees, and neither can read signs. Premapped data seems like the only way to address these blind intersections; the car must know what exists before it can see it. I am hoping that Tesla can use crowdsourced data from the fleet to generate this premapped data.
Again, not a blind stop sign. Warning sign, STOP-letter in the road. And then the sign, and the solid line across the road which is clearly visible long before the car starts to break.

It would be a lot easier if everyone would just acknowledge that the system is not on track to become autonomous this decade on existing hardware, if ever. Two years ago it had a DE rate of 5 miles/DE. Now It's like 15.
 
Secondly, this is not an occluded sign. There are large STOP letters painted on the road and there is a warning sign of the upcoming stop sign
...
It would be a lot easier if everyone would just acknowledge that the system is not on track to become autonomous this decade on existing hardware, if ever. Two years ago it had a DE rate of 5 miles/DE. Now It's like 15.
Tesla needs to actually read all signs, determine which are relevant, and drive accordingly. If they don't have the processing or memory to do that then admit it.

All they are doing is occasionally tacking on new display objects, and fudging around with % improvements to this or that behavior.

If they can't ever actually read the signs then there never will be any point. They can't even get the speed signs right after how many years.

Speed bumps were supposed to be detected, but it's hit and miss. Some road markings are detected but not always. Some car lights are detected but not always. And do they actually use that information, unclear.

If some problems are related to latency then I fail to see how they will ever manage to do all the hundreds of things they are still not doing? Are they just stringing this out because there's just no way?
 
  • Like
Reactions: spacecoin
Yes, Tesla cannot reliably see a stop-sign and stop without correct map and gps data even though there are multiple warnings signs and road paintings and the stop sign is visible for seconds. Now, they patched the map data. The elephant in the room is still here. What's new?
Yes. Pay no attention to the disease we're fixing the symptoms. The team continues to take cues from hospice/palliative care.
 
  • Like
Reactions: hooty