WILL TESLA EVENTUALLY KILL ME?

[Spartacus]

As a motorcycle owner this video is about autopilot being blind to certain motorcycles as it approaches from behind. This is, to say the least most concerning. Could be anecdotal evidence but the video poster has a good point about cameras being blind to distance in some cases. I know that statistically AP is safer but I for one don't want to be one of the few that slipped thru the statistical cracks. There is a good argument from Sandy Munro for "Forward looking infrared" (FLIR). Maybe there is a place for FLIR on the front of the car/truck? I once watched a video of an early radar equipped Tesla that stopped safely for a flock of ducks on the road in dense fog. The cameras could never pick this up nor would our own eyes.

Sponsored

 

[BillyGee]

Mine has never had a problem seeing riders, but I'm just one sample. The thing it really doesn't like is flatbeds with backwards cars on them.

When you're using autopilot you're supposed to be paying attention anyway in case it does make a poor decision. I always report when it does.

 

[CyberBC]

As a motorcycle owner this video is about autopilot being blind to certain motorcycles as it approaches from behind. This is, to say the least most concerning. Could be anecdotal evidence but the video poster has a good point about cameras being blind to distance in some cases. I know that statistically AP is safer but I for one don't want to be one of the few that slipped thru the statistical cracks. There is a good argument from Sandy Munro for "Forward looking infrared" (FLIR). Maybe there is a place for FLIR on the front of the car/truck? I once watched a video of an early radar equipped Tesla that stopped safely for a flock of ducks on the road in dense fog. The cameras could never pick this up nor would our own eyes.

This is sometimes called "inattention blindness". People are looking for cars so don't see motorcycles or bicycles. Expect it is worse for human drivers than AP. Here's a good article:

https://www.sciencedaily.com/releases/2018/01/180105082243.htm

I always assume I am not seen when riding.

 

[Dids]

37 seconds was what the time the computer had to decide the motorcycle was a problem.... If the motorcycle was going 65 (speed limit?) And the car was traveling @ 80 mph or 15 mph over the speed limit. Something about this scenerio doesn't make sense. The car had autopilot.... Maybe, seems people that crash claim it was on and it wasn't when an investigation is done. But say it was on and you are speeding along on autopilot and slowly catching up to a motorcycle that is going the speed limit. You would have to be inattentive for more than 37 seconds to not realize that you are catching up to a motorcycle that is traveling the speed limit.
Also I could not find where the first driver claimed his car was on AP, in fact news reports only say he was struck by a Tesla traveling behind him and that the accident is being investigated. The Utah driver did claim that his car was on AP but no news report on what investigation found.
I think the premise of this video is suspect. Yes 2 different Tesla's struck motorcycles and both cyclists died but that is about all that was reported. It is entirely possible that they wiped out and got hit.

 

[Cherokee180C]

Sorry Tesla Auto Pilot did not kill those Motorcyclist, the drivers did? What were they doing sleeping at the wheel? How come they did not see the motorcycle? I am sure autopilot will get much better as we continue down this path, but those idiot owners know 100% they are not level 5 autonomous cars. They should be charged with manslaughter.

 

[TheLastStarfighter]

Driving a motorcycle the odds are (relatively) high that something will kill you. The odds are (relatively) low it will be a Tesla.

 

[Spartacus]

Let me say that I'm a Tesla fanboi and have a CT3 on order. I also know it is more dangerous to go out on a motorcycle. I also defend ignorant statements by non EV people who repeat dumb things.
I too believe that you are not smart to buy any other brand, Tesla has the most advanced battery tech, superior software, better (mechanical) build quality, less is more engineering and proven superior reliability.

What I'm a little alarmed at is the tendency to defend Tesla at all cost. In response to this thread ... articles on inattentive drivers and the danger of M/C riding and statistics that it won't be a Tesla?

My point is maybe, just maybe there is a way to slightly improve something?
Put your defensive shields down for a minute and look at the facts. Are you aware that people drive while texting? There are those who actually have too much to drink, and then drive. Some people set Autopilot and put total reliance on "the car" when they should be paying attention. If cameras are at the 99% point why can't we have 100% with one FLIR system on the car to avoid instances like the flipped over trucks, stopped fire trucks, traffic cones to warn drivers and slower motorcycles?

Sandy on FLIR vs cameras, start at about 1:50 for the impatient.

 

[JBee]

I was never really a fan of the idea of getting rid of fusing radar with visual sensors. In my book they should scan everything that is useful and affordable, I mean they use ultrasonic all the time anyway. They could even use some of the SpaceX phased array tech from Starlink to do a solid state scanning Radar.

There's also a possibility that Teslas new deal with Samsung uses ToF cameras instead, that can measure distance to each pixel sensor with a global shutter, eliminating the need for scanning LIDAR and costly FLIR sensors to provide better distance whilst providing standard vision based feeds.

There are various threads discussing the options for and against.

What is the intention here, to cause alarm that there is a problem with motorbikes or certain lights, or discuss if radar should be reenabled? If the former, then I'd also have to point out that the cameras, and the distance between mounting positions on the Tesla itself will suffice to resolve range even to a single point or light source, especially so as it approaches. The ranging does not, to the best of my knowledge, use the distance between the lights of the other road vehicle at all to do this. It just doesn't need to, and makes it way more complicated than it has to be, if it at all it could reliably be done by assuming some distance between lights to begin with.

The video seems to be heading in the direction of general discreditation, but not by using any technical fact. More is simply not always better, and anyone who does sensor fusion will tell you this too.

 

[CyberOwl]

Stick to riding on the track. It’s more fun, and much safer

 

[Cyberman]

As a motorcycle owner this video is about autopilot being blind to certain motorcycles as it approaches from behind. This is, to say the least most concerning. Could be anecdotal evidence but the video poster has a good point about cameras being blind to distance in some cases. I know that statistically AP is safer but I for one don't want to be one of the few that slipped thru the statistical cracks. There is a good argument from Sandy Munro for "Forward looking infrared" (FLIR). Maybe there is a place for FLIR on the front of the car/truck? I once watched a video of an early radar equipped Tesla that stopped safely for a flock of ducks on the road in dense fog. The cameras could never pick this up nor would our own eyes.

To answer your question, yes. Tesla will probably kill you. But think about this: at the end of the day, if you're gonna be killed by something, wouldn't you rather it be by a Tesla?

 

[firsttruck]

....
My point is maybe, just maybe there is a way to slightly improve something?
Put your defensive shields down for a minute and look at the facts. Are you aware that people drive while texting? There are those who actually have too much to drink, and then drive. Some people set Autopilot and put total reliance on "the car" when they should be paying attention. If cameras are at the 99% point why can't we have 100% with one FLIR system on the car to avoid instances like the flipped over trucks, stopped fire trucks, traffic cones to warn drivers and slower motorcycles?

Sandy on FLIR vs cameras, start at about 1:50 for the impatient.

--------------

My early background is actually similar to Sandy's in that we both started our careers in about the same years and both started hands-on in industries where we personally hands on made stuff out of heavy duty steel. Differently than Sandy I moved to computers and software development after 3 years on machine shop floor.

I watch every Sandy Munro video because he has expertise in many things but he is not always best at everything. Munro has good team too but they also will not be experts in everything.

I think Munro team is correct that FLIR is good but Munro is not best at knowing sensor fusion & deep analysis of FSD in general, AI, Machine learning. Also most Munro experience with object recognition and sensing is with aircraft/missiles where in the sky you don't have to worry about how to detect it is a bridge and safe to go under or children vs squirrel.
Ground vehicle FSD is much harder and very different than automated aircraft obstacle avoidance.

For an independent analysis of FSD I listen to James Douma, Lex Friedman, Connecting the Dots, Comma ai | George Hotz.

I think Munro team are good but in regard to FSD they are no match to world leading team at Tesla or FSD teams of Waymo & GM Cruise. Waymo, GM Cruise and most other FSD competitors have almost everything & kitchen sink assortment of sensors yet few or maybe none are using FLIR.

 

[rr6013]

TeslaVISION is fundamentally aligned with human infallibility. Tesla are to be commended for a vision-only approach to safety. With a blended human and AI driver environment its a good thing that neither’s skill surpasses the other. Parity is a “good” even if it may not be equal. Assured safety will always be relative.

TeslaVISION exceeds human, can see better in low light and through mist-y conditions in addition to a wider field of view at higher resolution. All that technological improvement over human’s central vision limitation it still can’t match human peripheral perception, mental instinct and conditioned response. But FSD is working on perception. DoJo is set to tackle conditioned response. And FSD continues to improve, excel and still fail sometimes too.

Tesla is fixing flashing emergency vehicle collisions, phantom braking is resolving and drive thru motorcycles is the latest “runover” task for ML. The child one proved false data. TeslaVISION can fix errant, obvious and corner motorcycle cases.

Technologies have limitations, faults and errors just as humans have blind spots in vision, judgement errors and suffer from parallax.

Radar sees objects at night, in weather and shows movement across static backgrounds. But doesn’t see a semi-trailer stretched across the road if it's not moving either.​

Lidar remote senses and distance measures but as Elon pointed out 3D doesn’t work well blended and cannot transform into 2D fast enough to contribute to safety.​

​

TeslaVISION enables FSD to provide safer driving by A.I., ML and DOJO training over time it can become the highest embodiment of Level 4 automation.

But the system can never be perfect in a blended human-AI environment just safer.

 

[Dids]

--------------

My early background is actually similar to Sandy's in that we both started our careers in about the same years and both started hands-on in industries where we personally hands on made stuff out of heavy duty steel. Differently than Sandy I moved to computers and software development after 3 years on machine shop floor.

I watch every Sandy Munro video because he has expertise in many things but he is not always best at everything. Munro has good team too but they also will not be experts in everything.

I think Munro team is correct that FLIR is good but Munro is not best at knowing sensor fusion & deep analysis of FSD in general, AI, Machine learning. Also most Munro experience with object recognition and sensing is with aircraft/missiles where in the sky you don't have to worry about how to detect it is a bridge and safe to go under or children vs squirrel.
Ground vehicle FSD is much harder and very different than automated aircraft obstacle avoidance.

For an independent analysis of FSD I listen to James Douma, Lex Friedman, Connecting the Dots, Comma ai | George Hotz.

I think Munro team are good but in regard to FSD they are no match to world leading team at Tesla or FSD teams of Waymo & GM Cruise. Waymo, GM Cruise and most other FSD competitors have almost everything & kitchen sink assortment of sensors yet few or maybe none are using FLIR.

Agree with you! I worked with Teledyne for several years on sensor fusion for future force warrior. We ultimately dropped flir as it didn't add enough capabilities for the power and weight. Flir is best for detecting a thermal variant object, not so much for determining what that object is. NV was better at nuance. There is no way Tesla can't see enough with just cameras.
In other words it isn't a sensor problem. It is an AI problem determining what an object is and what to do about it.

 

[Spartacus]

Tesla is a great company but they should be challenged from time to time.
Something can be good, really good but can't it be better?
Tesla had a hard time fusing radar with cameras so they dropped radar. I still think this is temporary. They know as anyone does that cameras cannot see thru fog, heavy rain, blizzards, smoke ... no matter how good the cameras are. One day in the future Tesla will announce another sensing device (whatever that may be) that compliments cameras. Might take a better chip and software?

Radar can only see moving objects? Does this mean it will allow one to drive full speed into a concrete wall or parked truck?

 

[Dids]

Tesla is a great company but they should be challenged from time to time.
Something can be good, really good but can't it be better?
Tesla had a hard time fusing radar with cameras so they dropped radar. I still think this is temporary. They know as anyone does that cameras cannot see thru fog, heavy rain, blizzards, smoke ... no matter how good the cameras are. One day in the future Tesla will announce another sensing device (whatever that may be) that compliments cameras. Might take a better chip and software?

Radar can only see moving objects? Does this mean it will allow one to drive full speed into a concrete wall or parked truck?

Actually I don't know that cameras can't see through fog, heavy rain, blizzards and smoke. Do those obstructions uniquely absorb EM in visible light part of the spectrum vs the EM used by a radar? The answer is no.
They can see through skin and around corners.
https://www.newsweek.com/camera-see...thwestern-university-light-scattering-1651173

For that matter what makes a system a camera? Vs FLIR vs radar. It really is just the spectrum of EM that it is tuned to.

Sponsored