Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
Huawei ...

 
I have yet to see one example of those in the real world. And if they exist, it is likely that the car could angle itself to see around or that traffic in the street will have to choose a speed to be able to break for cars turning into the street both for humans and for computers drivers.
Here's an example from AI Day 2022 - Planning showing how FSD Beta handles temporary occlusions blocking the pillar camera:

planning occlusion.jpg


While this case has the other vehicle moving out of the way to restore visibility, it's not that uncommon for a very similar scenario to have that vehicle stop for traffic, e.g., entering a busy parking lot, or if you flip that vehicle around, it could be a vehicle creeping out to make a left turn, or if you pretend it was a static object like a fence or building when exiting a narrow street/alley.

To be clear, I agree Tesla will handle this with HW3 just like how humans need to deal with these scenarios such as waiting for visibility to clear or slowly creep out hoping cross traffic will stop or reroute with a right turn or avoid the intersection. Clearly some options aren't ideal and introduces some tradeoffs among safety, comfort and progress.
 
Here's an example from AI Day 2022 - Planning showing how FSD Beta handles temporary occlusions blocking the pillar camera:

View attachment 939541

While this case has the other vehicle moving out of the way to restore visibility, it's not that uncommon for a very similar scenario to have that vehicle stop for traffic, e.g., entering a busy parking lot, or if you flip that vehicle around, it could be a vehicle creeping out to make a left turn, or if you pretend it was a static object like a fence or building when exiting a narrow street/alley.

To be clear, I agree Tesla will handle this with HW3 just like how humans need to deal with these scenarios such as waiting for visibility to clear or slowly creep out hoping cross traffic will stop or reroute with a right turn or avoid the intersection. Clearly some options aren't ideal and introduces some tradeoffs among safety, comfort and progress.
Yeah, there are some scenarios that will require HW3 to either wait a bit extra or angle the car a bit awkwardly or maybe even in the extreme case just risk it and drive under the assumption that a car that can see it will decide to stop. Like this case:

But this is on roads that humans will struggle also, maybe not in all car types, but for sure in some car types. But it's not something that is making FSD impossible. Worst case is the waymo-solution of just not driving on those few roads before they are fixed...
 
  • Like
Reactions: jeewee3000
To be clear, I agree Tesla will handle this with HW3 just like how humans need to deal with these scenarios such as waiting for visibility to clear or slowly creep out hoping cross traffic will stop or reroute with a right turn or avoid the intersection. Clearly some options aren't ideal and introduces some tradeoffs among safety, comfort and progress.
One thing humans can do that FSD camera can't do is to bend forward for visibility - we have all done that.

Not sure why Tesla didn't put the side camera in the front or at least A pillar.
 
Not going to spoon feed you. Search in the AP forum. If you are too lazy you can live on your Hopium.

ps : Your profile says you are in Sweden. Either way you have zero experience with FSD, looks like.
Not with Tesla FSD, that's true, but I have been developing and been riding in another self driving car with a similar camera setup.

If you are not gonna contribute with anything else than ad hominem, then imo don't post or go to realtesla...
 
  • Disagree
  • Like
Reactions: STUtoday and EVNow
Not with Tesla FSD, that's true, but I have been developing and been riding in another self driving car with a similar camera setup.

If you are not gonna contribute with anything else than ad hominem, then imo don't post or go to realtesla...
Dude, I don’t need your approval to post. I won’t definitely spoon feed arrogant people for sure.
 
Last edited:
Is anyone familiar with how Tesla uses neural networks to train driving behaviors? What exactly is the output of the model? I assume the input is camera pictures, but the output of driving can be braking, turning, changing lanes, etc. How do you fit all of these into one neural network?
 
Is anyone familiar with how Tesla uses neural networks to train driving behaviors? What exactly is the output of the model? I assume the input is camera pictures, but the output of driving can be braking, turning, changing lanes, etc. How do you fit all of these into one neural network?
They don't do that. They have dozens of NNs doing particular things. They still do most of the planning and control using normal code.
 
This post on AP sub-forum I think exemplifies the first experience of FSD. Unless someone persists, they will just stop using FSD ...

Even as an ADAS, FSD needs to do better, much better, before the average Tesla user will want to use it (let alone pay for it).

People like us with a lot of hours as safety drivers, can do even majority of the drives with zero disengagements - but its currently a niche product.

ps : forgot to post the link.

 
Last edited:
This post on AP sub-forum I think exemplifies the first experience of FSD. Unless someone persists, they will just stop using FSD ...

Even as an ADAS, FSD needs to do better, much better, before the average Tesla user will want to use it (let alone pay for it).

People like us with a lot of hours as safety drivers, can do even majority of the drives with zero disengagements - but its currently a niche product.
But that one guy watched some Youtube videos...he's an expert. Our thousands of FSD Beta miles mean nothing...neither do people like Chuck Cook who has repeatedly provided examples of where the cameras cannot see properly and have to creep too far.
 
But that one guy watched some Youtube videos...he's an expert. Our thousands of FSD Beta miles mean nothing...neither do people like Chuck Cook who has repeatedly provided examples of where the cameras cannot see properly and have to creep too far.
ps : I didn't realize it was a sarcastic post ...

Did you see that post ? They are explaining their personal experience ...

Its the experience of someone just starting to use FSD that counts. Our experiences and YT doesn't mean much.

FSD is very jerky and seems very uncertain when starting to use first. It is one thing to see someone climb Mt Everest and something entirely different if we climb it.
 
Last edited:
  • Love
  • Like
Reactions: navguy12 and JRP3
Did you see that post ? They are explaining their personal experience ...

Its the experience of someone just starting to use FSD that counts. Our experiences and YT doesn't mean much.

FSD is very jerky and seems very uncertain when starting to use first. It is one thing to see someone climb Mt Everest and something entirely different if we climb it.
I was joking.
 
  • Funny
Reactions: EVNow
So, I am wondering about Elon's comments on FSD during the shareholder meeting. He said he's currently going days without safety-related interventions. That is certainly not my experience. Is that most likely because he's got a different build, or his interpretation of safety-related intervention vs mine? I think FSD is a great product showing great potential, but his representation today doesn't match my personal observations...
I reacted to Elon's comment too. A couple of things came to mind;
1. He didn't say how far he drives per day.
2. He didn't say how he defines a safety intervention. If he considers nothing less than an impending collision a safety intervention, then his statement makes sense. There might then have been non safety related interventions.

My interventions are sometimes when FSD beta is headed at 30 mph toward a 7" protruding man hole or a gaping pothole. If I were Elon, as product architect and techno king, I wouldn't consider that a safety intervention, just a fun test of how the undercarriage holds up.
 
Last edited: