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

How is FSD ever going to work with current camera layout?

This site may earn commission on affiliate links.
That said, what would the radar tell you? That there is a something ahead. So what action will be determined? Turn left, turn right or backup?
I'd hope it would start with "keep it between the white lines and don't hit anything" functionally. Otherwise stop or get help.

A few years back drove from Columbus OH to Bozeman MT in my MS. Coming westbound in I-90 is a posted 70 MPH and somewhat twisty road. It's dark, high gusty winds, and pouring rain. Road markings were barely adequate. If I were manipulating the controls, I would have stopped. But AP anyway was able follow the road. Finding a brave lead car also helps. It drove better than I could given the circumstances.

But I have no doubt Radar played a major role there. Perhaps Vision will work as well. Certainly Vision "sees" much better than I do but will it be good enough? Then again, I'm not sure an L5 car would be expected to be atonomous in ANY weather condition anyway
 
I'd hope it would start with "keep it between the white lines and don't hit anything" functionally. Otherwise stop or get help.

A few years back drove from Columbus OH to Bozeman MT in my MS. Coming westbound in I-90 is a posted 70 MPH and somewhat twisty road. It's dark, high gusty winds, and pouring rain. Road markings were barely adequate. If I were manipulating the controls, I would have stopped. But AP anyway was able follow the road. Finding a brave lead car also helps. It drove better than I could given the circumstances.

But I have no doubt Radar played a major role there. Perhaps Vision will work as well. Certainly Vision "sees" much better than I do but will it be good enough? Then again, I'm not sure an L5 car would be expected to be atonomous in ANY weather condition anyway
Radar cannot see the white lines 🤷🏽‍♂️
 
  • Like
Reactions: arghx7 and KArnold
Then again, I'm not sure an L5 car would be expected to be atonomous in ANY weather condition anyway
An L5 car would be expected to operate under any conditions where a human driven car would. That's a bit of a fuzzy criteria, but I don't think the standards are very precise in that area. Nonetheless, the car is not expected to be superhuman, so icy conditions, dense fog, extreme rain, etc. would, presumably, be cases where the L5 vehicle may not operate.
 
An L5 car would be expected to operate under any conditions where a human driven car would. That's a bit of a fuzzy criteria, but I don't think the standards are very precise in that area. Nonetheless, the car is not expected to be superhuman, so icy conditions, dense fog, extreme rain, etc. would, presumably, be cases where the L5 vehicle may not operate.
Until the vast majority of cars have similar capabilities, it would be unsafe for an AV to continue driving even with superhuman weather-penetrating sensors. It has to operate within the community of vehicles and drivers, and not be in the path or dash in and out of view of other weather-blinded operators who may still be on the road (advisedly or not).

I think about this all the time when I see people talk about radar being able to "see" in severe conditions. Aside from the technical overestimation of this capability, it just doesn't represent a realistic license to keep driving.
 
  • Like
Reactions: Boza and enemji
I am blown away by peoples denial here. I use it EVERY DAY. It does work in downtown, crazy busy, high volume pedestrian areas. It works in china town SF amazingly well. If works in windy mountain roads going 60mph extremely well. It might slow a little more than I’d like around curves down to around 45, but it successfully does it! (Where the speed limit is 50. So not a big deal)

How can people still think vision doesn’t work.

What are you people not seeing. I’m so confused.
I hope you mean Chinatown during the day, not at night :)

What I don’t see is working auto wipers and auto high beams - which prevents me from using the AP, let alone FSD.

I have been driving to/from Tahoe regularly, even in snowstorm. Few strange use cases (production FSD/AP with radar):
- Whiteout. I cannot see more than 10’. Somehow, the car visualizes the road perfectly.
- Because there is no exhaust/engine, the back/front of the car gets easily totally covered with snow. Until the pillar cameras get foggy, the FSD continues to work well - apparently, ignoring rear view. Somehow, USS work most of the time. Either they are not susceptible to a bit of snow or their position is such that the snow easily falls/does not accumulate.
- Heavy snow, super slippery. I would consider safe speed to be no more than 20-25mph. The thing wants to go with the speed limit (65mph).

If you notice the pattern here is inconsistency, compared to a human driver. Certainly far, far away from autonomous.
 
Last edited:
What you describe in the first paragraph is 100% how it works in a Level 0 to potentially Level 4 ADAS, although at Level 4 there need to be safeguards in place that allow the vehicle to safely pass ownership of the driving task back to the human or safely pull over and stop — that’s another spot where redundant sensors could be important.

Now let’s think about true Robotaxis and potentially vehicles without steering wheels or pedals and how they’d need to handle this with a total lack of any controls for a human. Or if your vehicle is supposed to be out driving around autonomously, shuttling people from Point A to B to C, and something like this happens even if it has controls. And the risk being assumed by companies who now own the driving task while the vehicle is operating autonomously. I don’t think you’re passing the driving task off to customers using the vehicle.

There are good reasons driverless Robotaxi services are currently operating where they operate, in arid climates like Arizona, Nevada, Texas.
If a scenario happens where even a human shouldn’t be driving, why would we expect an autonomous vehicle to persist? Shouldn’t it also pull over with flashers and wait out the conditions until safe to resume, as a human does?
 
  • Like
Reactions: JHCCAZ
I don’t think that’s the bar people are looking to be met. The goals for autonomy seem to be beating humans at safe driving, not sport driving. Even before reaching autonomy we are seeing better safety numbers and less loss of life with ADAS and FSD Beta engaged. Far fewer collisions per million miles driven than those driven manually, and improving consistently as it matures.

That’s not to say no one could work on a rally driving focused system. But it would have to be trained for that from the ground up. Just like a human would. If you stick a daily commuter in a rally race it’s not going to go well. It may be easier than general city driving since the scope of “rally driving” is far more narrow. Much like how narrow AI is by definition far easier than AGI (general) to achieve. This is why Waymo can operate in a small geofenced area and FSD is operating anywhere there are roads, and taking much longer to develop.
 
U better find that quote
That's pretty trivial, because he says it A LOT. Here's a quote from 2020:


He's backed away from using SAE autonomy levels in his predictions, pivoting towards fuzzy, ill-defined goals like "safer than a human":


Nobody takes his pronouncements on this seriously any more.
 
If you are referring to Tesla’s numbers - they are highly misleading. They picked a special use case (where FSD can be enabled) and compare it to the entire population of human use cases.

FSD Beta can be enabled everywhere I’ve tried it other than some parking lots where it didn’t want to engage. The numbers they report are not FSD, just autopilot and navigate on AP. So far FSD has zero crashes, so crashes per million miles is just zero. That will be more meaningful when it’s rolled out to millions rather than just the current 160,000. Especially if it were to rain zero. Seems unlikely since that would be insane. But welcome for sure.
 
So far FSD has zero crashes, so crashes per million miles is just zero.
They aren't saying zero crashes on FSD Beta are they? We've seen crashes posted before. TeslaRaj for one, AI Addict, Dennis Perez, and that guy who ended up in the field.

Those are all crashes involving FSD Beta and had some damage to car or property. They don't even have to be caused by FSD, just involving it. Other companies have to report incidents of any type, with any incidental damage or injury.
 
Go on YouTube and watch some drive videos of FSD Beta testers (TESLA EV owners). It is remarkable—not perfect, but it is working incredibly well on both highways and city streets. Remember: FSD does NOT need to be perfect; only better than a human driver and have notably fewer accidents/incidents. Not here there yet, but is is fascinating to see how far FSD has come in just handful of years.
 
  • Like
Reactions: bjhinkle