jhogan2424
Well-known member
- Thread starter
- #91
Sandy Munro posted a video on YouTube today and he said QUOTE “Tesla giving up on radar is goofy”
Sponsored
It just goes to show that as knowledgable as Sandy Munro is he is not infallible. I guess I would want to see what he meant by "goofy" before going further. I heard him talking more than once (this morning with Chris from Dirty Tesla for example) about needing more cameras at intersections and at other times about better maps because of unmarked roads, and both of these tell me that he doesn't understand the technology that Tesla is using. Maybe he doesn't understand the strengths and weaknesses of various sensor combinations. I don't know, but that is enough for me to discount the "goofy" statement.Sandy Munro posted a video on YouTube today and he said QUOTE “Tesla giving up on radar is goofy”
nice pic!What about Llamas in the fog? Can it see Llamas in the Fog?
Agreed. I respect Sandy's engineering skills but he doesn't understand much about FSD... the comment about the cameras on the traffic lights... <cringe>It just goes to show that as knowledgable as Sandy Munro is he is not infallible. I guess I would want to see what he meant by "goofy" before going further. I heard him talking more than once (this morning with Chris from Dirty Tesla for example) about needing more cameras at intersections and at other times about better maps because of unmarked roads, and both of these tell me that he doesn't understand the technology that Tesla is using. Maybe he doesn't understand the strengths and weaknesses of various sensor combinations. I don't know, but that is enough for me to discount the "goofy" statement.
I have not seen (as you may have) an in-depth explanation from Musk or Karpathy (or anyone else) about what the technical reasons for dropping radar were. We have experienced the phantom braking issues ourselves so I used that as an example. As a result, one form of cost is risk to the driver from not perceiving the conditions correctly.Yes. That's an important part of system design. The AI guy recognizes that the sensor integration is flawed. He then goes on to say that the single sensor is so good that it's npt worth bringing the system design up to snuff.
That is almost never true. What is true is that the additional sensor, or sensor type, comes with a cost (it must be integrated properly and a cross checking integrity monitoring system must be part of that) and a benefit. If the OEM does not feel the benefit is worth the cost, then he abandons the second sensor. That's the case here.
The video posted earlier in this thread is why I want radar. Some here would lead you to believe the radar had nothing to do with the car stopping but they are wrong. It was radar, and radar only, that told the car to brake. Cameras had zero to do with the car braking in that particular situation.I have not seen (as you may have) an in-depth explanation from Musk or Karpathy (or anyone else) about what the technical reasons for dropping radar were. We have experienced the phantom braking issues ourselves so I used that as an example. As a result, one form of cost is risk to the driver from not perceiving the conditions correctly.
I am guessing that it is very difficult to get enough training examples where the radar and vision systems disagree to be able to choose radar over vision. If this is what is going on then the cost (in time) might be too high to continue using radar.
I have written I don't know how many programs where results are ranked and we only considered those in the top N positions fro further analysis. If radar is always in the top 2 positions but never first (or at least rarely) then it is effectively being ignored already. This seem as likely a scenario as any. That is another kind of cost.
We can add up the effects of all the different kinds of cost functions, and of course we can keep coming up with more examples that illustrate one cost function or another, but we still don't know which, if any, of these was relevant in Tesla's decision.
As mentioned, if AutoPilot cannot resolve what is going on then it kicks out and tells the human to take over. Then it is on us. Perhaps that is really the currently best solution. I watched several of Dirty Tesla's videos of FSD Beta last winter and it did quite well on snowy roads with no markings, but tended to drive too far to the right and he felt uncomfortable with it. I was too. So even without radar the autopilot system has to slow down and drive differently in different conditions.
Can someone enumerate the various scenarios that make them want radar and tells us what the ramiifications of false positives between radar and vision would be on those scenarios?
Your vision is very good at estimating acceleration, range, etc. Look at what is involved in throwing a ball for someone to catch, or catching one thrown at you - your eyeballs and brain are processing a tonne of spatial and velocity data.No, actually it isn't. Try an experiment. I'll be particularly interested to hear how good you are at estimating acceleration.
FSD is better because it has 8 cameras as opposed to my two and can compute faster than I can. But it is fascinating that even 2 cameras is enough to feed the Kalman filter that is clearly running in our brains in the background. Now my wife doesn't have one or if she does it isn't tuned properly and it is fascinating (and terrifying) to observe the way she drives.
Do you have some evidence of some sort? I certainly don’t know what exactly the answer is but short of some direct evidence one way or another, it’s just one stance against another.The video posted earlier in this thread is why I want radar. Some here would lead you to believe the radar had nothing to do with the car stopping but they are wrong. It was radar, and radar only, that told the car to brake. Cameras had zero to do with the car braking in that particular situation.
You don't know this was triggered by radar alone this is a huge assumption on your part.The video posted earlier in this thread is why I want radar. Some here would lead you to believe the radar had nothing to do with the car stopping but they are wrong. It was radar, and radar only, that told the car to brake. Cameras had zero to do with the car braking in that particular situation.
My evidence is the video. Just watch it. No room for interpretation. There is no hidden agenda in the video, just use common sense when watching it.Do you have some evidence of some sort? I certainly don’t know what exactly the answer is but short of some direct evidence one way or another, it’s just one stance against another.
Evidence would almost have to be some indication on the center console or some logs of the event. Otherwise we have no idea how the car made the decisions it made and what evidence it was working off of.
Yes Sir I will concede that it is an assumption but it is an assumption derived from only a few possibilities that in the end only leave one plausible explanation. It’s ironic that so many here are rallying around the video only Teslavision and proclaiming how much better video is than radar but when an actual video is presented as evidence they suddenly don’t believe video is good enough and can’t be trusted. Video is good enough for Teslavision but video isn’t good enough when offered as evidence.You don't know this was triggered by radar alone this is a huge assumption on your part.
There are other videos (in this thread) which clearly show the Tesla with vision only is capable of seeing cars 2 or 3 cars in front of the vehicle with vision only.
What you suggest "braking with radar alone" is not how Tesla's safety features work. They take all of the inputs available and combine them into one model of the world around them, then the car reacts based on that model. This is what Musk is talking about in his tweet above when he refers to combining inputs.
I believe the video shows the car braking early.Yes Sir I will concede that it is an assumption but it is an assumption derived from only a few possibilities that in the end only leave one plausible explanation. It’s ironic that so many here are rallying around the video only Teslavision and proclaiming how much better video is than radar but when an actual video is presented as evidence they suddenly don’t believe video is good enough and can’t be trusted. Video is good enough for Teslavision but video isn’t good enough when offered as evidence.
This is not that complicated. The answer is because the stopped vehicles can not be seen visually. Can’t be seen by me, can’t be seen by you, cant be seen by the occupants, can’t be seen by any of the Tesla’s video cameras, there are no visual clues for any cameras to pick up on. There is no visual data concerning the stopped cars for the Tesla to process. However, there is radar data to be processed. That is how it can be deduced that radar was certainly the primary reason and probably the only reason the car braked. If presented to a competent jury to consider this video would be proof beyond a reasonable doubt that radar alone braked that car. It could not be any more cut and dry in my opinion. If this video does not make the case that radar has value than nothing will change your minds. It really doesn’t matter in the end because Tesla is dropping the radar no matter what a young hillbilly from Arkansas believes. The sore thumbs I have from typing so many replies on this subject are for nothing because my opinions will never reach Mr. Musk and wouldn’t matter to him if they did. You guys win this one, I gotta tap out.I believe the video shows the car braking early.
I don't see anything in that video that suggests remotely *WHY* it starts braking early. It is just as likely it's braking due to input from one of the other 8 cameras on the car. We've seen a video where a Tesla sees beyond the car in front of it with no trouble using vision only.
Exactly, what makes you so certain that the Tesla in your video must be using radar as opposed to vision?
When I watch the video in slow motion, I can see a brake light from the 3rd vehicle ahead, followed by the edge of the 2nd vehicle just to the left of the vehicle immediately ahead, followed by the brake light of the 2nd vehicle. All that happens before the audible alarm goes off. I assume that Tesla is using as much data as possible from the cameras possibly even NIR that doesn’t even get shown in the video. Are the visual cues the factor that set off the alert? I have no idea. Was radar somehow able to detect the impending accident first? I have no idea what the radar data looks like so who knows. I do know that the computer is MUUUUCH faster than I am at seeing what is happening and responding to the circumstances. So if I can see possible signs of the accident before the alarm goes off then there is a considerable chance that what I saw was enough data for the car to set off the alarm.My evidence is the video. Just watch it. No room for interpretation. There is no hidden agenda in the video, just use common sense when watching it.
In the video, you can see brake lights ahead. It could have seen the stopping cars.My evidence is the video. Just watch it. No room for interpretation. There is no hidden agenda in the video, just use common sense when watching it.