Separate names with a comma.
Discussion in 'Model X' started by SoCal Tsla, May 28, 2017.
I'd rather plow through an empty cardboard box than have AEB engage....
Okay, now who knows the specifics about at what speeds and other conditions AEB is supposed to engage? What materials?
This might not be what's happening here, but there was an old Star Trek episode where they tested a robot with an AI to see if it would stop a simulated problem that would kill it if it were real. It didn't. Later they discovered it didn't fail the test, it saw right through it, and knew there was no real danger. A bad AEB will not see a box and run right through it. A decent/good AEB will see a box and stop. A really good AEB will see a box and run right through it. Now I doubt the Tesla AEB is really that good, but depending on how it detects imminent collisions, it might not detect certain material combinations. This is an additional consideration beyond just verifying other constraints like at what speeds it's active, and it may only be designed to prevent automobile collisions presently.
That said, it would be interesting to see a full suite of tests in different conditions be performed against multiple different car brands that claim to have automatic emergency braking.
Note that this is not an easy feature to perfect. Make it too sensitive, and the car brakes randomly all the time. Make it too insensitive, and it doesn't engage when it needs to. I've dealt with similar scenarios before in programming. It's a difficult dance to perfect.
I doubt the radar will see a cardboard box, and although the camera would see it, I doubt the software is trained to recognize that shape.
If I were doing this test, I'd look at making a soft object that looks like the rear of another car, and is radar reflective.
Tesla's New Automatic Emergency Braking System Limited to 28 MPH
On a non-Tesla (Cadillac), my wife was backing up, and would have brushed against a rose bush if she kept going. It slammed on the brakes in reverse at low speed and scared the crap out of her.
But was that a solid enough threat to warrant AEB? It probably could not tell what kind of object it was, but did not set off the Pedestrian Warning System so it knew it wasn't a person. It does ID humans behind the car as humans.
Perhaps the AEB in the Tesla is looking for a stronger signal reflection than a box would allow. I would have put foil inside or outside of the box.
But kudos for doing the test. Some folk are in deathly fear of something stiffer than a cotton diaper touching their car.
For the artifact I'd start with a 4' x 8' x 1" sheet of foam from Home Depot ($12), put foil on one side (either), and hold it up with string a duct tape to supports. A lot more work, but will mimic a car reflection better. Maybe even 2 red circles of paper to mimic tail lights.
I'm wondering about AEB too.
Just look at this incident on a Dutch highway May 15th. https://s3.amazonaws.com/media.sergekroes.com/media/EVT_2017_05_15_07_23_06_F.MP4
Read more about it here.
AEB is radar based. Radar can only see things which reflect RF energy. Plastic and cardboard does not. You need to cover it in aluminum foil. Pull up to it slowly first and make sure TAAC sees it and you get a car symbol on the dash.
You knocked off this exact test from a year ago, which was full of the same comments:
This is what you need to do. Make sure you see the box "car" displayed on the dash. And keep in mind this is one of those difficult stationary obstacle detections that don't always seem to work.
You need to use proper target. For instance Finnish magazine "TM" uses visually car imitating target which also has radar reflecting material. For radar reflection they used radar reflectors whis are designed for sail boats. Some aluminium folio might also work.
Plastimo Radar Reflectors
AEB probably needs both visual and radar signal.
Talviauto 2017: Automaattinen hätäjarrutus
It can't just rely on video processing due to fog, hills at night, etc.
They have cars with Low Speed Automatic Braking that will not stop a collision all the time, but they will hit the brakes before impact to tighten the belts and reduce the impact to the 'low velocity' airbag or no-airbags impact settings. High velocity airbag deployment still kills people every year (benefits outweigh harm), but low speed deployment is far safer, very, very few deaths.
Trivia, airbags used to kill a significant number of people, and permanently cripple thousands. This was a gov't mandate that took years before the gov't allowed 2 speed airbags. NHTSA killed at least 290 people by dragging their feet.
I think everyone said it already: Your experiment only proves that Tesla does not detect PAPER. If there's no DETECTED obstacle, the department of AEB doesn't have to respond!
1) Failure to detect:
The system needs detect an obstacle and you can see that result on your Instrument Cluster Display as a car icon.
Your Instrument Cluster Display never did display any obstacles except in a fraction of a second for the gigantic Ditch Digger from far away.
At this stage, you need to figure out how your Tesla can detect a box before you can carry on the experiment.
Both camera and radar are responsible to detect an obstacle.
I don't know why the camera didn't detect the box. It could be that Tesla Vision needs some learning.
We know that radar can go through fog and a carton box so may be you can wrap it with aluminum foil.
2) Instrument Cluster Displays an obstacle--TACC:
Once you get your box detected as displayed on Instrument Cluster due to aluminum foil, you can try to drive toward the box with TACC or autosteer (if available) set at 18MPH.
The car should brake itself as if it encounters stopped cars at a red light.
3) Instrument Cluster Displays an obstacle--AEB:
If it works for TACC, it should work for AEB. Now, you can manually drive your car slowly toward the box.
How about we get a TMC representative to sit on the engineering team that designed the AEB system and report back to us customers. Why would we ever trust Tesla. We should question all of their claims. Why stop with AEB? Let's challenge...say...side collision. No...wait...how about steering.
People are like 80% water, right?
So what happens if you try to run your Tesla into a snow man? (It'll need a carrot and a hat to compensate for the difference in water content, though.)
Pedestrian collision warning is based on image recognition, since it is based on MobilEyeQ3.
Here's a link for how the Euro NCAP AEB test target (also used by IIHS) was designed:
This is how the test target looks relative to the actual vehicle.
Under the cover, there is foil to simulate the radar signature of a car:
This is the NCAP pedestrian test target:
AEB Pedestrian | Euro NCAP
A cardboard box is not going to cut it as a test target.
That is a good suggestion. Communications, demonstrations, lecturing, explanation would promote trust.
As amateurs, resources might be scarce so I wouldn't expect many testings.
Yeah I wish to see a test like that too
True, I'm considering putting some water balloons in there...
I don't think the shape matters, but yeah I should make it more radar reflective, will try that later lol