At what point does FSD start assisting?

[John K]

My use, I expect daily assistance in my freeway commute. Expect to use from onramp to offramp. Or maybe, when I cross into the carpool lane to the exit ramp. I don’t know if the system will be aggressive enough to get to the first carpool opening.

Autodrive will do most of what I need, the rest is available in FSD current state.

Advertisement

 

[moonrover]

This says a lot of positive about your driving skills while traveling thru a hostel environment.
Maybe we should customize the paint on your current vehicles for higher visibility and add the unbelievably loud and glass shattering air horn.
Or just a cow guard like a locomotive. You would win .... just saying
Haha

 

[Dids]

Cameras can see better than eyes. Algorithms can calculate angle and trajectory better than brain. AI already has better spatial awareness than human /time. Humans drop attention all the time and if nothing happens no one notices. If something happens people say well they are only human. Many edge cases happen while human is driving that go unnoticed and since driver didnt notice it's as if it never happened. We as riders need to be ok with FSD occasionally killing us, we certainly are ok with it when we ride on roads with human drivers.

 

[Saskateam]

I am ok with FSD. I am far more likely to kill myself driving then a computer and I am a good driver (said like all humans who make mistakes while driving). I have some of the features in my current vehicle like adaptive cruise control and lane keep assist. I use it all the time and driving is way more relaxing and it has kicked in a lot.

 

[Jhodgesatmb]

Cameras can see better than eyes. Algorithms can calculate angle and trajectory better than brain. AI already has better spatial awareness than human /time. Humans drop attention all the time and if nothing happens no one notices. If something happens people say well they are only human. Many edge cases happen while human is driving that go unnoticed and since driver didnt notice it's as if it never happened. We as riders need to be ok with FSD occasionally killing us, we certainly are ok with it when we ride on roads with human drivers.

Cameras see differently than human eyes, that doesn’t make them better. Cameras require humans to convert what they record into something that can be interpreted. I am happy to acknowledge what ML can do with massive amounts of information, but I don’t think that is enough, yet. Perhaps with a thousandfold improvement in current technology but that won’t happen in the next five or ten years. Also, you can get greater field of view with a camera than with the human eye but all we have to do is turn our head a bit and we seamlessly take in, integrate, evaluate, and respond to everything. To say that people don’t pay attention, etc., is really irrelevant because we all agree on that and we are talking about self driving cars and how to improve them. I am a believer but I also want more ‘understanding’ from my self driving car. I don’t accept a powerful bit-processor as being enough, so if I suggest a way to improve it is with that backdrop in mind.

 

[Dids]

Cameras see differently than human eyes, that doesn’t make them better. Cameras require humans to convert what they record into something that can be interpreted. I am happy to acknowledge what ML can do with massive amounts of information, but I don’t think that is enough, yet. Perhaps with a thousandfold improvement in current technology but that won’t happen in the next five or ten years. Also, you can get greater field of view with a camera than with the human eye but all we have to do is turn our head a bit and we seamlessly take in, integrate, evaluate, and respond to everything. To say that people don’t pay attention, etc., is really irrelevant because we all agree on that and we are talking about self driving cars and how to improve them. I am a believer but I also want more ‘understanding’ from my self driving car. I don’t accept a powerful bit-processor as being enough, so if I suggest a way to improve it is with that backdrop in mind.

If a vision system can see smaller less visible things in more conditions it is proper to call it better and it is incorrect that cameras require humans to interpret (this is so patently false that it's a ridiculous statement). The human brain only pays attention to the vision area covered by your thumb at arms length approximately and rapidly shifts that attention unless it is taken up by other task. My point was that we accept this human limitation as safe and made an argument against requiring 100% perfection in AI when we are so willing to accept less from humans. Am I personally willing to accept this... No! Should I be willing? That is the discussion in this thread.

 

[DMC-81]

I don't know when FSD kicks in, but Tesla should definitely hire these Stanford students. I'd like to have a "Cyber Khana" mode on my new stainless steel wonder :


(By the way, the DeLorean is electric.)

 

[Saskateam]

I hope infrared is also included in the AI information available. Bring in a HUD on the windshield that would make the heat visible would help the driver to act on situations that are on the fringes of what AI can currently handle and our reactions would help ML to improve.

 

[ajdelange]

FSD is going to assist with crash avoidance for pedestrians, cyclists, trashcans and such. Does anyone know what size an object has to be for a Tesla to react and brake/avoid? Will it aggressively stop for a squirrel, bouncing ball, or small dog? What if a bird flies in front of the vehicle? I have only test driven a Model X and have a CT on order so I am pretty curious.

When you say "size" do you mean radar cross section, accoustic cross section or physical size? There are three sensors looking forward and any of them can detect an object in front of the car. But detection is only part of the problem. The thing detected must be classified as either a threat or not and the cat does this by analyzing the signals from the sensors. The current system is infamous for declaring overpasses as threats and for missing semi trailers. What would it do with tumbleweeds and squirrels? I don't know. But I do know that no camera, radar or accoustic sensor suite will ever be as good as a pair of human eyes in classifying threats correctly.

 

[Dids]

When you say "size" do you mean radar cross section, accoustic cross section or physical size? There are three sensors looking forward and any of them can detect an object in front of the car. But detection is only part of the problem. The thing detected must be classified as either a threat or not and the cat does this by analyzing the signals from the sensors. The current system is infamous for declaring overpasses as threats and for missing semi trailers. What would it do with tumbleweeds and squirrels? I don't know. But I do know that no camera, radar or accoustic sensor suite will ever be as good as a pair of human eyes in classifying threats correctly.

Cant the just use 3d camera and a time of flight camera. Do they need radar? I guess but seems redundant. The acoustic is just for close by objects I guess?

 

[PACEMD]

FSD is going to assist with crash avoidance for pedestrians, cyclists, trashcans and such. Does anyone know what size an object has to be for a Tesla to react and brake/avoid? Will it aggressively stop for a squirrel, bouncing ball, or small dog? What if a bird flies in front of the vehicle? I have only test driven a Model X and have a CT on order so I am pretty curious.

Here is a video reflecting just how rudimentary some of the detection features currently are. I don't really care about a bouncing ball in a parking lot, how about a deer or a drunk pedestrian/bicyclist on the edge of the highway?

 

[ajdelange]

Radar has the tremendous advantage of being able to see through dust, fog and rain and the secondary advantage of being able to measure distance to a target fairly accurately. Yes, the accoustic sensors are for short range.

 

[PeteDCNT]

I can wait for FSD the same way I could not wait to afford my Ti-100 calculator. But I fear the lessons learned was a generation that grew up on technology, for many lost the ability to simple math in their heads. At what point will the driver have the ability to take control if divers down the road lose their ”skill” to drive.

 

[Dids]

I can wait for FSD the same way I could not wait to afford my Ti-100 calculator. But I fear the lessons learned was a generation that grew up on technology, for many lost the ability to simple math in their heads. At what point will the driver have the ability to take control if divers down the road lose their ”skill” to drive.

I have never believed that using a tool makes you dumber. While knowing how to use a calculator does mean you stop doing simple math in your head it also means you know how to do complex operations that the average person who is great at simple math can't comprehend. The professor argues that if you use a calculator how are you going to build a bridge if things change and calculators become unavailable is foolish. If things change so badly I need to know how to grow mushrooms not how to build a bridge. Really wish they taught me how to use MatLab instead of teaching me how to do calculus.

 

[ajdelange]

Thing is that anyone who knows calculus can easily learn to use Matlab but the reverse path is more difficult. I used to marvel at the young engineers who could do all kinds of things with Matlab but they didn't really know what they were doing. IOW they could develop an algorithm that worked but they couldn't tell you how it worked or why it worked. I used to infuriate the management by refusing to use what I called "canned software". This is going to get too philosophical fast.