Does Tesla have a big math problem?

[SanDiegoM3]

Starting at the other side of this. What is the human reaction to "side view"
Regular humans have a reaction time of around .250 (1/4 second) but that doesn't include the time it takes to recognize the situation then hit the brake/go pedal etc.
Within that 1/4 second one camera at 36fps would have 9 whole frames to figure out what's going on and react, or 72 frames for all of them.
Add the fact the drivers spend most of their time looking forwards and many seem to need extra prodding to bother looking sideways or backwards.
On the face of it, it would seem the computer would react significantly faster than a human and spot oncoming cars from either direction faster than a human (who can't look in both directions at once)

I can absolutely see further and clearer than the side cameras. Without a doubt. I can also predict better than the software can currently. I can plot what lane oncoming traffic is. I can preposition myself to be successful. You’re overly simplifying complex visuals, thinking, prediction, and maneuvering.

If you think otherwise then let’s bet $1,000 that Tesla will launch either more or higher resolution cameras looking to the side in the next 5 years.

 

[Twiglett]

main point of the comment was that in terms of straight frame rate and field of view and all the following comments went along with that.
But the computer has the human beaten hands down. It can react faster, see all directions. The only thing it can't do is see further - but that wasn't the basis of the thread.
It was all about not being able to react quickly enough within the distance that it could see - but it obviously CAN react quickly enough.
So just because 1 frame is 3 feet means nothing. It just sounds a lot, but if it only needs 2-3 frames (6-9 feet) then its got plenty of time.
A human looking one direction has no way to see in the other direction at the same time, something that the computer can do easily.
A human is also really good it remembering and correlating, so you can look one way, then other and back again and retain your environment very well. But a human has to be because you can on see in one direction.
The computer doesn't need to do that.
Sure it can screw up, but most of the time it's got nothing to with frame rates and resolution.

 

[Dan D.]

main point of the comment was that in terms of straight frame rate and field of view and all the following comments went along with that.
But the computer has the human beaten hands down. It can react faster, see all directions. The only thing it can't do is see further - but that wasn't the basis of the thread.
It was all about not being able to react quickly enough within the distance that it could see - but it obviously CAN react quickly enough.
So just because 1 frame is 3 feet means nothing. It just sounds a lot, but if it only needs 2-3 frames (6-9 feet) then its got plenty of time.
A human looking one direction has no way to see in the other direction at the same time, something that the computer can do easily.
A human is also really good it remembering and correlating, so you can look one way, then other and back again and retain your environment very well. But a human has to be because you can on see in one direction.
The computer doesn't need to do that.
Sure it can screw up, but most of the time it's got nothing to with frame rates and resolution.

In some tricky situations I have asked the passenger to help by looking for traffic too. Never done that? "It's ok my way".
I may not always have a passenger but if I do, nothing wrong with having a second set of eyes. That still counts.

I can see people on the street reacting to something like a speeding car about to run a light. Humans can notice things that aren't obvious to a computer.

I can look rearward in the mirror and still recognize activity through the windshield with peripheral vision. Not with clarity, just noticing movement and lights - but we CAN see in multiple directions simultaneously. You give us far too little credit.

 

[Twiglett]

In some tricky situations I have asked the passenger to help by looking for traffic too. Never done that? "It's ok my way".
I may not always have a passenger but if I do, nothing wrong with having a second set of eyes. That still counts.

I can see people on the street reacting to something like a speeding car about to run a light. Humans can notice things that aren't obvious to a computer.

I can look rearward in the mirror and still recognize activity through the windshield with peripheral vision. Not with clarity, just noticing movement and lights - but we CAN see in multiple directions simultaneously. You give us far too little credit.

All true and agreed...
Certainly at least for some people, driving is something to pay attention to.
For others though, they've been looking right at me, looked away, looked back right at me and STILL pulled out in front of me
Or they've been so busy on the phone or chatting to a passenger or just not bother to look.
Some humans can do it and do it extremely well - others shouldn't be in charge of a vehicle but they only had to pass their driving test once before they wreak a lifetime of havoc on other road users, only avoiding accidents because of the skill of other road users.
Most are somewhere in the middle, we all hope we are better than the middle group though

 

[drtimhill]

In some tricky situations I have asked the passenger to help by looking for traffic too. Never done that? "It's ok my way".
I may not always have a passenger but if I do, nothing wrong with having a second set of eyes. That still counts.

I can see people on the street reacting to something like a speeding car about to run a light. Humans can notice things that aren't obvious to a computer.

I can look rearward in the mirror and still recognize activity through the windshield with peripheral vision. Not with clarity, just noticing movement and lights - but we CAN see in multiple directions simultaneously. You give us far too little credit.

But the point is most accidents are not caused by lack of perception, they are caused be lapses of attention, or distraction (passengers, phones etc), or incapacitation (alcohol, tiredness) or just plain stupidity (road rage, brake checking). Things that an AV doesnt suffer from. So even if you are right, and the car is less capable than a human from a perception standpoint, you have to ask if that disadvantage is offset by the continuous unyielding vigilance of the car. And let's not forget cameras and NNs are improving every day. Are humans?

Oh, and there is already a video on YouTube of FSD beta reacting to (and dealing with) a red-light runner before the human driver even noticed.

 

[AndreP]

But the point is most accidents are not caused by lack of perception, they are caused be lapses of attention, or distraction (passengers, phones etc), or incapacitation (alcohol, tiredness) or just plain stupidity (road rage, brake checking). Things that an AV doesnt suffer from. So even if you are right, and the car is less capable than a human from a perception standpoint, you have to ask if that disadvantage is offset by the continuous unyielding vigilance of the car. And let's not forget cameras and NNs are improving every day. Are humans?

Oh, and there is already a video on YouTube of FSD beta reacting to (and dealing with) a red-light runner before the human driver even noticed.

Human ability isn’t improving every day but the technology is, and I’m not referring to AVs.

We’ll start seeing more vehicles rolling off the production lines with systems that are designed to detect distraction, intoxication, fatigue, medical emergencies, etc and take action accordingly.

We don’t necessarily need to replace human drivers to have a big impact on accident rates, there are many smaller and easier things that can be done to reduce the risk of stuff we know contribute to most accidents. Just taking a cloth strap and fastening it across your torso can make a world of difference in terms of injury/fatality risk.

If you look at areas of the world with the highest rates of car accidents/injury/fatality per capita, they’re poor countries with substandard infrastructure and a population driving old bare bones vehicles or mopeds and such. Having good infrastructure alone makes a huge difference.

Even when it comes to autonomous vehicles, we might be able to achieve comparable safety benefits by merely supplementing human ability with good, solid emergency braking and other driver-assist functions.

 

[EVNow]

If you think otherwise then let’s bet $1,000 that Tesla will launch either more or higher resolution cameras looking to the side in the next 5 years.

Let’s bet that those cameras will be far worse than human visual acuity. How much ?

 

[drtimhill]

We’ll start seeing more vehicles rolling off the production lines with systems that are designed to detect distraction, intoxication, fatigue, medical emergencies, etc and take action accordingly.

We don’t necessarily need to replace human drivers to have a big impact on accident rates, there are many smaller and easier things that can be done to reduce the risk of stuff we know contribute to most accidents. Just taking a cloth strap and fastening it across your torso can make a world of difference in terms of injury/fatality risk.

If you look at areas of the world with the highest rates of car accidents/injury/fatality per capita, they’re poor countries with substandard infrastructure and a population driving old bare bones vehicles or mopeds and such. Having good infrastructure alone makes a huge difference.

Even when it comes to autonomous vehicles, we might be able to achieve comparable safety benefits by merely supplementing human ability with good, solid emergency braking and other driver-assist functions.

That may be true, but at some point the car is spending so much time looking over your shoulder and taking over in more and more complex situations that you might just as well let it take over driving most of the time. Instead of trying to get you out of a mess, you simple avoid getting into the mess in the first place.

We are moving from accident mitigation (crumple zones, seat belts, air bags), to accident avoidance (emergency braking, lane departure prevention), then to accident elimination (preemptive control, predictive paths etc). And each stage is going to move moe and more driving responsibility over to the car.

 

[AndreP]

That may be true, but at some point the car is spending so much time looking over your shoulder and taking over in more and more complex situations that you might just as well let it take over driving most of the time. Instead of trying to get you out of a mess, you simple avoid getting into the mess in the first place.

We are moving from accident mitigation (crumple zones, seat belts, air bags), to accident avoidance (emergency braking, lane departure prevention), then to accident elimination (preemptive control, predictive paths etc). And each stage is going to move moe and more driving responsibility over to the car.

Truly self-driving cars aren't coming any time soon though, these are things we can do now to drive down the bad numbers. By the time something like Level 5 is here, it's safe to say accident rates will already be significantly lower.

 

[drtimhill]

Truly self-driving cars aren't coming any time soon though, these are things we can do now to drive down the bad numbers. By the time something like Level 5 is here, it's safe to say accident rates will already be significantly lower.

You dont need L5 to realize this, L2/L3 is just fine to get enough of the benefits.

You should also be wary of the "safety lamp" effect. In the 18th century a "safety lamp" was invented to make coal mining safer. It prevented explosions in the mines. Was the result lower deaths from explosion? No. Because the cynical mine owners reasoned "with the safety lamp we can now send the men to much more dangerous parts of the mine!".

Adding a safety feature can make people think they can drive more dangerously/fast/stupidly .. result, more accidents not fewer.

 

[SanDiegoM3]

Let’s bet that those cameras will be far worse than human visual acuity. How much ?

That’s not the bet. Elon and Tesla engineers said the existing vehicle is fully capable of full self driving at the robo taxi level without an attendant.

The latest update is objectively worse than some of the past versions that were more timid but didn’t create such risky situations.

People severely underestimate how hard driving is in a general sense rather than location specific.

 

[SanDiegoM3]

That’s not the bet. Elon and Tesla engineers said the existing vehicle is fully capable of full self driving at the robo taxi level without an attendant.

The latest update is objectively worse than some of the past versions that were more timid but didn’t create such risky situations.

People severely underestimate how hard driving is in a general sense rather than location specific.

 

[qdeathstar]

Objectively worse? I don’t think so. It is a lot better in many areas.

 

[StealthP3D]

Objectively worse? I don’t think so. It is a lot better in many areas.

True. FSD keeps getting gradually better. From a user perspective it may seem like slow progress but it's actually happening quite quickly and will stun the world when it's mature by appearing to happen almost instantly. One minute there was no autonomous driving, the next minute it's everywhere!

 

[drtimhill]

The latest update is objectively worse than some of the past versions that were more timid but didn’t create such risky situations.

I'm curious how you figured out it was "objectively" worse?

 

[EVNow]

That’s not the bet.

Why not ? Because you know whatever camera they can add, it will always be orders of magnitude worse than human eye ?

The latest update is objectively worse than some of the past versions that were more timid but didn’t create such risky situations.

Based on some weird ULTs Chuck tries ? Do you have FSD Beta ?

To me it is worse as I've posted elsewhere .... apparently not for others.

 

[SanDiegoM3]

Why not ? Because you know whatever camera they can add, it will always be orders of magnitude worse than human eye ?


Based on some weird ULTs Chuck tries ? Do you have FSD Beta ?

To me it is worse as I've posted elsewhere .... apparently not for others.

It's not the bet because Tesla said the car was FSD ready. A MILLION robot axis on the road in 2021! Anyone with a brain could tell the side cameras were a severe weak point with a wide angle 720p resolution camera, it only took me a month or two of ownership in 2019 to figure that out. It simply doesn't have enough visual acuity to differentiate objects from background noise at the required distances. Not even counting the bad positioning of the B-pillar camera. A wide angle side camera doesn't need to be 8k resolution, but certainly can't be 720p at 36fps with no other side camera to help with stereo vision.

I have a FSD car, but I have refused to update because of the V11 UI fiasco and my current version is fine. When everyone else finishes beta testing FSD I'll jump over, but I probably won't have the car by then.

 

[SanDiegoM3]

I'm curious how you figured out it was "objectively" worse?

I watched the video with the car pulling into the road unsafely, leaving itself stuck in the road (not median) as it waited for oncoming traffic to pass - but forgetting it could get annihilated staying in the road like it did, etc. Did you even watch the video?

 

[SanDiegoM3]

I watched the video with the car pulling into the road unsafely, leaving itself stuck in the road (not median) as it waited for oncoming traffic to pass - but forgetting it could get annihilated staying in the road like it did, etc. Did you even watch the video?

Even random users who aren't in this thread come across the same obvious flaws. It's odd that people defend a design flaw, as if Tesla is infallible.

FSD Beta 10.12

This version is still unusable in NYC. FSD simply can't figure out a path with 10+ cars in an intersection and dozens of pedestrians. Id love to see how Tesla will solve this one. If they want robo taxies they will need them in the two major markets (ie LA/NYC). They are not even attempting to...

teslamotorsclub.com

 

[Yelobird]

It's not the bet because Tesla said the car was FSD ready. A MILLION robot axis on the road in 2021! Anyone with a brain could tell the side cameras were a severe weak point with a wide angle 720p resolution camera, it only took me a month or two of ownership in 2019 to figure that out. It simply doesn't have enough visual acuity to differentiate objects from background noise at the required distances. Not even counting the bad positioning of the B-pillar camera. A wide angle side camera doesn't need to be 8k resolution, but certainly can't be 720p at 36fps with no other side camera to help with stereo vision.

I have a FSD car, but I have refused to update because of the V11 UI fiasco and my current version is fine. When everyone else finishes beta testing FSD I'll jump over, but I probably won't have the car by then.

Lol. It’s objectively worse is your “anyone with a brain” review yet you don’t Have beta, have not Used it, and Choose to be a dozen software updates behind because your on strike agains the v11 UI. Lol. Your input has zero value sorry.