Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Model X Crash on US-101 (Mountain View, CA)

This site may earn commission on affiliate links.
Car camera + voice nav could query your attentiveness...
"Car: Quick, what color is the car in front of us?"
"You: Uh, um, let's see... red,.... no blue..."
"Car: You clearly weren't paying attention. You must take over full control now. Bye."

It's a WHITE Tesla!

ERROR ERROR ERROR. The correct color is 'Pearl White'

Autosteer Unavailable for the Rest of This Drive

You clearly weren't paying attention. You must take over full control now. Bye.
 
Chapter 2 - Older Driver Highway Design Handbook, January 1998 - FHWA-RD-97-135
25.gif


Back in the "good old days" when they used to make them safer ?

II. Interchanges - Older Driver Highway Design Handbook, January 1998 - FHWA-RD-97-135
 
...
Any other ideas?

Sure, here are a few, limited to this accident in particular, rather than the other legitimate observations posted here that should also be addressed.

1) Use the rear camera view of the lane line to "parity check" the lane line information provided by the front cameras to make the product more robust to glare. One set of cameras should be glare free. The lane definition that goes under the front of the car should be the same lane definition that comes out from under the back of the car.

2) Check periodicity of dashes and cross lines (chevrons). Anything with a frequency component should be avoided.

3) Recognize diverging lines as the tip of a gore area, and draw them in the screen as no-go areas. You don't have to record every cross hatch. The gore areas could be diplayed as solid keep outs. Anything that looks like the tip of a bull's horn to the side of the road get's no-goed.

4) Use high definition map data to help with 3) construction of no-go zones.

5) Alarm wide lanes as points of 'use confusion' and noticeably slow the vehicle.

These are all actionable, require no new sensors, and can be stream processed from the video images.

Looking at the relative motion of an object with respect to the pavement underneath it can tell you if it is moving. There should be a vector field produced by looking at the position of distinctive features in two successive frames. Moving objects will show a discontinuity of motion vectors with respect to the stationary objects around them.
 
Last edited by a moderator:
What advertising? Tesla does not do any that I know of.

Since they do not own the messaging around their cars (advertising is done for them; by everyone else, the media, social media), Tesla doesn't have the opportunity to properly position Autopilot. And I feel strongly that Tesla should own that message.

How do they not own the messaging? The message is quite clear on their website. The message comes from Elon Musk who heads up their 'big reveals' and does
interviews, goes on tv, etc etc. The best advertising any business can have is word of mouth, Tesla has plenty of that. They may not spend a billion dollars advertising
and making TV commercials, but they certainly own the messaging more than most companies.
 
Sure, here are a few, limited to this accident in particular, rather than the other legitimate observations posted here that should also be addressed.

1) Use the rear camera view of the lane line to "parity check" the lane line information provided by the front cameras to make the product more robust to glare. One set of cameras should be glare free. The lane definition that goes under the front of the car should be the same lane definition that comes out from under the back of the car.

2) Check periodicity of dashes and cross lines (chevrons). Anything with a frequency component should be avoided.

3) Recognize diverging lines as the tip of a gore area, and draw them in the screen as no-go areas. You don't have to record every cross hatch. The gore areas could be diplayed as solid keep outs. Anything that looks like the tip of a bull's horn to the side of the road get's no-goed.

4) Use high definition map data to help with 3) construction of no-go zones.

5) Alarm wide lanes as points of 'use confusion' and noticeably slow the vehicle.

These are all actionable, require no new sensors, and can be stream processed from the video images.

Looking at the relative motion of an object with respect to the pavement underneath it can tell you if it is moving. There should be a vector field produced by looking at the position of distinctive features in two successive frames. Moving objects will show a discontinuity of motion vectors with respect to the stationary objects around them.
Interesting and potentially useful concepts, thanks.
 
  • Helpful
Reactions: 22522
As a FYI, You will want to look up lane splitting. It's not legal in your state so you may not have encountered it. It's not the same as lane sharing.

If a lane splitting motorcycle is going up on my right, my "assumption" is Tesla moves over to the left if the left is clear.

If it's not, it holds position. I'm not testing it though. Autosteer goes off for me in this scenario.

Neither AP1 nor AP2 currently detect lane-splitting motorcycles based on my experience during various commutes home over the past two years.

However, I find that having AP engaged while driving in the HOV lane–even in stop-and-go traffic–makes it much easier for me to watch for motorcycles coming up on my Model X because I'm not worrying about drifting out of the lane. This in turn lets me move over earlier to give motorcycles more room when they pass.
 
  • Like
  • Informative
Reactions: bhzmark and MXWing
So the issue/ 'bug' is that you commanded the car to change lanes and it did? Were you expecting it to ignore you due to the single white line which has no legal significance in many states?

I thought the larger issue that video from @cbdream99 showed was that Autopilot would use the right-hand lane marker inside the gore area as part of the driving lane, but it was hard to tell from the video whether it was actually following that, or still trying to complete the lane change assuming both lanes were still going straight.

Either way @cbdream99, please do not continue the experiment. No need to put yourself in danger.
 
  • Like
Reactions: mongo
Neither AP1 nor AP2 currently detect lane-splitting motorcycles based on my experience during various commutes home over the past two years.

However, I find that having AP engaged while driving in the HOV lane–even in stop-and-go traffic–makes it much easier for me to watch for motorcycles coming up on my Model X because I'm not worrying about drifting out of the lane. This in turn lets me move over earlier to give motorcycles more room when they pass.
AP1 used to “latch on” to the lane-splitting motorcycles in early versions of the sortware. It would cause my car to accelerate a little until it reverted back to tracking the leading car, after the motorcycle passed it.
It does nothing of the sort anymore. It displays the motorcycle on the screen as it passes, but no reaction.
One reason why I'd like AP to hold closer to the left alone when in the far-left lane. Agreed that I can be much more pro active and see motorcycles coming, and steer left to give them plenty of room.

Wow, this is off-topic. Hard to tell by now, but yup! I'm calling myself out.
 
  • Like
Reactions: ddkilzer
And if it finds too many lines, it can get very confused.

To me, it looks like the paint on the left-hand lane marker is also faded (right after the crosswalk) and Autopilot may have lost the lane marker, but you could be right that the crosswalk may have also confused it.

Looks like you have a vehicle with AP2 based on the purchase date. I'm slightly curious which software update you had installed at the time as well since (in my opinion) navigation on "city streets" noticeably improved in 2018.10.4, although "wide-lane" support is a double-edged sword (helps in some cases, but still rough/needs fine tuning in others).
 
Others might not feel the same, but I'd rather drivers treat my motorcycles like they would any other vehicle. Don't run into me, but don't weave when I approach either. Lanes are 12' wide, if you just stay within 1 foot from the center, we can drive a small car through a normal gap between 2 Camrys. Only 2 pickups with trailer mirrors are an issue, not sedans.
 
Slowly s
I thought the larger issue that video from @cbdream99 showed was that Autopilot would use the right-hand lane marker inside the gore area as part of the driving lane, but it was hard to tell from the video whether it was actually following that, or still trying to complete the lane change assuming both lanes were still going straight.

Either way @cbdream99, please do not continue the experiment. No need to put yourself in danger.

I watched a couple times to see if it was doing that, but it seemed like @cbdream99 was already canceling the maneuver before the gore right lane was established. Mobile is not the best platform for frame by frame checks, so I could be mistaken.
Regardless, I fully agree with you on the latter sentiment, be safe out there people!
 
  • Like
Reactions: ddkilzer
If someone is interested in seeing how the CT6 with super cruise handles this...I can ask my buddy in the bay area to mess with his new Caddy.

That would be fascinating.

We can place bets.

My bet is that it will work just fine despite the low sun during the morning.

Cadillac probably falls back on HD maps like Tesla should do.

Of course I'll lose the bet if GM doesn't have this road white listed.
 
I thought the larger issue that video from @cbdream99 showed was that Autopilot would use the right-hand lane marker inside the gore area as part of the driving lane, but it was hard to tell from the video whether it was actually following that, or still trying to complete the lane change assuming both lanes were still going straight.

Either way @cbdream99, please do not continue the experiment. No need to put yourself in danger.

Thanks, I am ready to brake at any moment and make sure safe around. in any case once it goes into the non lane area, I immediately brake and steer it back.

The lanes are close to straight. You bring up good point, it could have trying to complete a lane change too but once it got into the non-lane area, I won't assume anything and take immediate correction, AP is good tool but I still will keep full attention on the road and see if anything is wrong.

Interesting thing is long before the exit on any solid right lane line (similar to what you seen after the correction in the video), it recognizes the shoulder or something else and won't try to change lane if the blinker is on.
 
  • Like
Reactions: ddkilzer and mongo
The markings of the road make a huge difference. I watched the videos that were made, illustrating how AP will take you of course, describing they had to take over. If you watch closely one line is very faded or non extent so the AP recognizes our sees the darker more prominent line and takes that line or road marking as a main road and centers to it. I have observed in my own driving that it will (ap) get lost and drift over a little then corrects it self and recenters. If I am driving up and over a steep hill or an aggressive turn it will lose it self and drift or swerve trying to find the center again. I realise that that is a limitation to the system and plan to take over control. AP has limitations agreed but it has saved me multiple times from accidents that other drivers cause.
 
Dan Noyes of ABC7 News has another follow-up report involving a similar accident six months ago in a model s.

I-TEAM EXCLUSIVE: Tesla crash in September showed similarities to fatal Mountain View accident

Looked up the street view for CA-92 at the gore point with the Hesperian Exit and saw this. The accident with the red MS in this I-Team Report said it happened last September. This google map view is from October 2017 so doubt it had changed much in one month's time. Thankfully this MS driver had the benefit of a properly extended crash attentuator barrier and lived. Given the driver was driving against morning rush hour traffic, he likely was driving at or near highway speeds upon impact. Unfortunately the I-Team report doesn't give us as many details as we here would like to know. For example did he see the barrier at the last minute and try to take over?

92atHesperianExit - 1.jpg


The report doesn't indicate which lane the Red MS was driving in (on 92 or in the exit lane for Hesperian) but I could see if he was in the left most Hesperian exit lane how his AP could have missed picking up on the broken solid white left line in the exit lane and instead picked up on the solid white that runs on 92 and tried centering on that. In this case he impacted the barrier on the right side of his car and ended up on the left side of the barrier. Of course there's also the possibility that the driver was in the 92 right lane and being blinded by the sun didn't see his exit signage or barrier until the last minute, signalled to turn right attempting to move over to the right to take the exit but hit the barrier before he could safely do so. We've seen the car will cross the white line on a turn signal and then apparently from what I understand lock on to the gore lines as if they were lane lines.

So bad lane markings for AP combined with glare from the morning sun were factors as well. At this point drivers using AP should understand why Tesla said you need to stay aware of the road at all times. Not much you can do about driving in a direction with a rising or setting sun and I honestly don't know if road lane markings will universally be kept in good enough condition to rely on all the time. Certainly measures like painted gore lane chevrons (if we could count on them being painted and maintained, or rumble strips) would help prevent or avoid these situations. Don't know if AP recognizes painted chevrons as no go - not a factor in either the Mt View or Hayward crash. The rumble strips if placed far enough back in the gore area would certainly get the attention of the driver.

Here's two screen shots from the ABC7news I-Team report of the Hayward's MS impact and the truck driver's dashcam view as he approached the barrier.

92atHesperianExitMS - 1.jpg

92atHesperianExitSun - 1.jpg
 
Last edited: