"Blind Spot Detection is disappointing"

[animorph]

I don't see anything in the current specs or user manual (for the X) that say blind spot detection is a feature of my car. The first warning in the user manual for auto lane change is:

"Warning: It is the driver's responsibility to determine whether a lane change is safe and appropriate. Auto Lane Change can not detect oncoming traffic in the target lane, especially fast moving vehicles from the rear. Therefore, before initiating a lane change, always check blind spots, lane markings, and the surrounding roadway to confirm it is safe and appropriate to move into the target lane."

I know it was promised on older cars, and this thread was started long ago, but people seem to be expecting a feature that isn't there (with apologies to the older car owners who were promised BSM).

I waited for AP2 to get a competent BSM using the side/rear cameras. Still waiting, but eventually it will get here. In the meantime I'm checking the mirrors and turning my head to check the blind spots myself.

 

[clostridium]

It also doesn't work as well in the dark, especially once the headlights of the vehicle next to you are past the side camera. Basically, in the dark camera can only see lights, not the whole car, which makes tracking what's next to you harder.

Using cameras combined with ultrasonic sensors makes that not much on an issue. For cars directly next to your vehicle where the headlights aren't visible to the cameras the ultrasonic sensors work fairly well. For vehicles in other locations around the car the cameras would be fine.

Looking at the big picture - I don't see how there will ever be a successful autonomously driving Tesla vehicle if they can't demonstrate a functional blind spot detection system. Eventually an autonomous car will have to change lanes. So if the sensor suite on the car now is inadequate for a blind spot system then we know the autonomous driving plan is not achievable with this sensor suite.

 

[oktane]

BSM works well on any modern car, minus Tesla.

 

[whitex]

Using cameras combined with ultrasonic sensors makes that not much on an issue. For cars directly next to your vehicle where the headlights aren't visible to the cameras the ultrasonic sensors work fairly well. For vehicles in other locations around the car the cameras would be fine.

Do you have any data to support that? Two inferior systems are surely going to work better than either one, but this is safety critical stuff, it needs to work every time, not 95% of the time (vs. 50% today).

 

[electracity]

The camera system should be able to tag any car at any angle. Assuming AP2 hardware is good enough to advance to at least level 3.

Correctly modeling all vehicle surrounding the car is a fairly modest subset of FSD.

 

[boaterva]

The camera system should be able to tag any car at any angle. Assuming AP2 hardware is good enough to advance to at least level 3.

Correctly modeling all vehicle surrounding the car is a fairly modest subset of FSD.

Exactly. How is the car supposed to drive itself if it can't see what's in the blind spots reliably 24 hours a day? I'd think it needs to work for us as well as for itself!

 

[whitex]

The camera system should be able to tag any car at any angle. Assuming AP2 hardware is good enough to advance to at least level 3.

Correctly modeling all vehicle surrounding the car is a fairly modest subset of FSD.

Camera works on light. If it's dark and the lights on the car are out of frame, the entire frame is dark, no distinction between car and no car. Also, with just non-stereo camera, there is always the challenge of distinguishing between car in next lane and car 2 lanes over - hard problem in the light, but gets much, much, harder to do when all you have to work with are 2 dots of headlights. Even humans have a problem there if you just show them pictures. Radar on the other hand is simple but effective in figuring out whether there is something near you.

 

[electracity]

Camera works on light. If it's dark and the lights on the car are out of frame, the entire frame is dark, no distinction between car and no car. Also, with just non-stereo camera, there is always the challenge of distinguishing between car in next lane and car 2 lanes over - hard problem in the light, but gets much, much, harder to do when all you have to work with are 2 dots of headlights. Even humans have a problem there if you just show them pictures. Radar on the other hand is simple but effective in figuring out whether there is something near you.

Fortunately there is always light outdoors. Vehicles are also emitting infrared.

Plus, the car doesn't start looking at the new lane when a lane change is desired. It is (or will be ) mapping all cars within sensor range and anticipating future position of all vehicles. It's going to know that the blurry blob in the blind spot is a car because it tracked that vehicle to that location.

Or so is the plan.

 

[whitex]

Fortunately there is always light outdoors. Vehicles are also emitting infrared.

Plus, the car doesn't start looking at the new lane when a lane change is desired. It is (or will be ) mapping all cars within sensor range and anticipating future position of all vehicles. It's going to know that the blurry blob in the blind spot is a car because it tracked that vehicle to that location.

Or so is the plan.

Yea, I get the general idea of the plan, but considering automatic wipers are still not working, I am not holding my breath on the rest of the stuff. When I got our AP2 car, I didn't buy into EAP/FSD hype but thought to myself "at least the safety features will be better than the AP1 car". I'm not so sure of that anymore, so far it isn't.

 

[alcibiades]

It's absolutely mind-boggling that our EAP camera suite cannot do blind-spot monitoring. Whether Tesla promised the feature or not, the lack of it should be embarrassing to them.

 

[whitex]

It's absolutely mind-boggling that our EAP camera suite cannot do blind-spot monitoring. Whether Tesla promised the feature or not, the lack of it should be embarrassing to them.

That's because it is easy to put cameras in the car connected to a computer. The hard part is to write the software. AP2 is a little like x.com from Elon - ok, x.com started with a little bit more functionality than AP2 when it was released .

 

[alcibiades]

That's because it is easy to put cameras in the car connected to a computer. The hard part is to write the software. AP2 is a little like x.com from Elon - ok, x.com started with a little bit more functionality than AP2 when it was released .

For sure. Alas, Tesla seems left with the software writers no one else wants.

 

[clostridium]

Do you have any data to support that? Two inferior systems are surely going to work better than either one, but this is safety critical stuff, it needs to work every time, not 95% of the time (vs. 50% today).

As others have noted, the system isn't starting with a vehicle just sitting right next to it, vehicles move in and out of the blind spots. The car can track with ultrasonic and camera sensors and would then know if there is something sitting right next to it even in the dark.

There are blind spot monitoring systems that work with just cameras - in fact the first one ever was in a Volvo and was camera only - so while I don't have data there are other companies that have demonstrated that this is a viable solution. I can't imagine that adding in the ultrasonic sensor data would make the sensing worse.

 

[whitex]

As others have noted, the system isn't starting with a vehicle just sitting right next to it, vehicles move in and out of the blind spots. The car can track with ultrasonic and camera sensors and would then know if there is something sitting right next to it even in the dark.

There are blind spot monitoring systems that work with just cameras - in fact the first one ever was in a Volvo and was camera only - so while I don't have data there are other companies that have demonstrated that this is a viable solution. I can't imagine that adding in the ultrasonic sensor data would make the sensing worse.

Adding ultrasonic would only make it worse if it had a lot of false, which today it doesn't (ok, curbs show up as warnings, but that's a good false, you don't want to change lanes into a curb). That said, you say volvo's gen1 was camera but next gen was not. Why did they switch? If their camera based system worked well, I suspect they wouldn't change to a different one.

Anyhow, bottom line is today Tesla's BSM is a joke. Someday, maybe, possibly, AP2 cars might get a better one, though it may or may not require FSD hardware upgrade, so $10K cost for those who just want BSM. I think the title of this thread summarizes the situation very well.

 

[AnxietyRanger]

There are blind spot monitoring systems that work with just cameras - in fact the first one ever was in a Volvo and was camera only

You are right about the Volvo system, but it was also much slower than the radar based systems where radars are in the rear bumper. Volvo also discontinued the visual BLIS later in favor of radar. Volvo now has two radars in the rear bumper as most of the rest of the industry...

The original Volvo BLIS also had one major flaw that the Tesla system also has - it can not see towards the sides (much) if being blocked by a vehicle next to it, so the side camera is useless as a rear cross-trafic alert sensor.

IMO one of the least sensible decisions was to ship AP2 without rear corner radars. I really think those would have been useful additions and that there is also some photographic prototype info suggesting Tesla was testing them early on on the Model X (possibly with a Level 3 aiming EyeQ "AP 1.5").

Fast-approaching cars are the area where radars especially help, if the car is not yet in the blind spot but approaching it fast, perhaps somewhat blocked by a trailing car... radar could help see those. Radar really is the best for that. Especially useful in the German autobahn...

 

[clostridium]

You are right about the Volvo system, but it was also much slower than the radar based systems where radars are in the rear bumper. Volvo also discontinued the visual BLIS later in favor of radar. Volvo now has two radars in the rear bumper as most of the rest of the industry...

The original Volvo BLIS also had one major flaw that the Tesla system also has - it can not see towards the sides (much) if being blocked by a vehicle next to it, so the side camera is useless as a rear cross-trafic alert sensor.

IMO one of the least sensible decisions was to ship AP2 without rear corner radars. I really think those would have been useful additions and that there is also some photographic prototype info suggesting Tesla was testing them early on on the Model X (possibly with a Level 3 aiming EyeQ "AP 1.5").

Fast-approaching cars are the area where radars especially help, if the car is not yet in the blind spot but approaching it fast, perhaps somewhat blocked by a trailing car... radar could help see those. Radar really is the best for that. Especially useful in the German autobahn...

The Volvo system used less cameras - one on each of the side view mirrors pointed backwards. I also do not believe that it was tracking objects but rather just saying "is something there at this instant." It was also hampered by significantly less computing power than what is available today vs. 2004. The hope is that a combination of tracking objects, the greater number of cameras (including not just the additional side cameras but also the additional front and rear) combined with ultrasonic sensors will allow the system to work. Which it has to if they want to do FSD.

It sounds like it is technically feasible however it's important to point out that Tesla hasn't demonstrated that it will work yet which is discouraging to say the least. Only time will tell.

 

[AnxietyRanger]

The Volvo system used less cameras - one on each of the side view mirrors pointed backwards. I also do not believe that it was tracking objects but rather just saying "is something there at this instant."

Yes, of course. I do remember the system - you could clearly see those camera bulges on the side mirrors and even the warning lights were visible outside of the car.

AP2 has the added benefit of the rear-view camera that could be used for cross-traffic monitoring when backing up. I was at the moment thinking in terms of the immediate system, the EAP, which is said to use four cameras - so those side-marker cameras from BSD. They can be easily blocked from seeing far and wide if a car or building alongside the car blocks the view...

I agree, only time will tell. I will add one thing, though: I do believe Tesla can make a functional BSD on the AP2, using the cameras. But I do think with rear corner radar additions they could have made a better one.

We shall see.

 

[whitex]

I'm curious, who are those 10% of

AP2 has the added benefit of the rear-view camera that could be used for cross-traffic monitoring when backing up. I was at the moment thinking in terms of the immediate system, the EAP, which is said to use four cameras - so those side-marker cameras from BSD. They can be easily blocked from seeing far and wide if a car or building alongside the car blocks the view...

Camera's can also be blinded by the headlight or the sun, radar on the other hand... but why are we debating this? Tesla BSM as it stands today is completely unreliable and chances are Tesla BSM will not get there till AP3 or AP4. Sadly for AP2, it's just another experiment like AP1 - once superseded by the next gen, it will be relegated to "old platform nobody has time for", like AP1 today.

 

[AnxietyRanger]

Camera's can also be blinded by the headlight or the sun, radar on the other hand... but why are we debating this? Tesla BSM as it stands today is completely unreliable and chances are Tesla BSM will not get there till AP3 or AP4. Sadly for AP2, it's just another experiment like AP1 - once superseded by the next gen, it will be relegated to "old platform nobody has time for", like AP1 today.

I guess at least myself am not debating anything on this thread, just pondering out loud.

I agree the ultrasonic-based blind spot detection, be it on AP1 or AP2, is currently pretty much useless. But I acknowledge that on AP2 it does have the potential to get much better once they start using the cameras for it. Whether or not it will be sufficiently good or as good as radars, remains to be seen.

This, of course, returns to the whole philosophy of self-driving sensors, where Tesla is going against the grain, so to speak, compared to the rest of the industry. Rest seem to be looking at 360 radars and lidar in addition to 360 visual, while Telsa - so far - is set on only radar and 360 visual as their stated philosophy. (All have 360 ultrasonics also, of course.)

That may change for AP3, of course, who knows.

 

[FooBar]

An easy improvement is to copy what many Honda models do: when the driver flips the signal indictor show the the rear (or better the side) camera view on half of the display. Tesla has a way to show rear camera on half the screen, but it is all time or none at all. While not as good as the radar based systems, something I could live with.