-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
New feature: Autosteer Stop Light Warning
- Thread starter strangecosmos
- Start date
Most people think i hate Tesla but i'm don't i think its a good thing, they got a traffic light detection that works close to 50% of the time and 10% of the time in the day and released it. They are forerunners.
I applaud them for that.
So it only works 50% of the time at night and 0% in the day. FSD should be around the corner... 
Try again during the night time. I hear the text appears in the release notes only 10% of the time during the day...
(In seriousness EU could be excluded for now from the fun.)
Nick's Life
Tesla Life
Hey, I was finally able to capture the Autosteer Stop Light Warning!!! Took a lot of tries and not exactly sure why it worked on this one intersection, but I got it!
ahkahn
Active Member
diplomat33
Average guy who loves autonomous vehicles
At the 1 mn 30 mark, Autopilot STOPS at an intersection with a red light on its own. Feature is in early access, not available yet.
Last edited:
diplomat33
Average guy who loves autonomous vehicles
Most people think i hate Tesla but i'm don't i think its a good thing, they got a traffic light detection that works close to 50% of the time and 10% of the time in the day and released it. They are forerunners.
I applaud them for that.
Most people think I hate bladder but I don't. He was able to make a trollish argument based on a small sample size. I applaud him for that. Nice try!
Based on what everyone is saying. Its hit and miss.
What this proves (like we needed any further proof.. ) is that elon was BS ing like always about coast to coast by end of 2017. Then double downing triple downing, saying end 2017 or latest early 2018, then every time he was asked it was in 3 months.
diplomat33
Average guy who loves autonomous vehicles
Did you see the video I posted from early access where autopilot is stopping at red lights on its own? Apparently the user was able to turn on the functionality to stop at red lights in the 2019.8 update:
It proves that EAP is able to see red lights and stop for them at busy intersections. So, Tesla's FSD development can handle red lights and stopping at intersections pretty reliably already based on that video. The reason you think it is hit and miss is because the warning does not always come on for every single red light. That's because the feature is set to only give the warning when it thinks the car will blow through a red light. So if the car is going too slow or slowing down, it won't give the warning since it won't meet the criteria for "blowing through a red light". Maybe Tesla needs to adjust that threshold? But the feature is working as intended. And based on the video of EAP smoothly stopping at red lights at intersections, I would say Tesla is making good progress on FSD.
Last edited:
diplomat33
Average guy who loves autonomous vehicles
If you read the release notes, it is clear that "autosteer stop light warning" does not give warnings just for seeing a red light, it only gives warnings when it thinks autopilot is at risk of blowing through a red light. That is a key distinction. It is why it does not always give a warning at a red light. It's misleading because it gives the impression that AP is not detecting red lights when it actually is.
@verygreen is there a campaign for a driver stopping at red lights on AP or anything related to red lights?
Did you even watch the video? That is what you call reliably? running a red light? Mis-categorizing red lights multiple times in just a handful of attempts, including missing traffic lights completely?
Then this video of stop signs: Recognizes and stops good on first stop sign, recognizes but stops late on second stop sign (dangerously crossing the intersection), doesn't recognize the third stop sign and completely blows past it.
PRETTY RELIABLY?
Cool the fanboy spin. It's not close to reliable but it's nice to see that they are actually working on it and could deliver something in like 3 months maybe, 6 months definitely.
Good gravy, it's the early release of a WARNING system. If it catches any red lights, it is doing better than 99.99999% of cars out there that have ZERO red light warning systems.
diplomat33
Average guy who loves autonomous vehicles
Did you even watch the video? That is what you call reliably? running a red light? Mis-categorizing red lights multiple times in just a handful of attempts, including missing traffic lights completely?
Then this video of stop signs: Recognizes and stops good on first stop sign, recognizes but stops late on second stop sign (dangerously crossing the intersection), doesn't recognize the third stop sign and completely blows past it.
PRETTY RELIABLY?
Yes, I watched the video. Let me walk you through it:
0:20 mark: correctly detects red light
0:50 mark: correctly detects red light
1:20 mark: correctly detects red light
1:34 mark: correctly detects red light
2:22 mark: correctly detects red light
2:47 mark: correctly detects red light
3:21 mark: correctly detects green light.
3:51 mark: correctly detects red light
4:52 mark: correctly detects change to green light
5:18 mark: misidentifies red as green and then corrects itself
6:33 mark: correctly detects green light
6:48 mark: correctly detects a green light at first but then thinks its red
7:13 mark: does not detect green light
7:30 mark: correctly detects green light
7:52 mark: correctly detects green light
Again, not perfect. Get's 12 out of 15 right to my count. That's 80% right. Not good enough for full public release which is why it is still in early access. But for an alpha, early access, feature, it looks pretty good.
Its not early access or even alpha. Its dev. VG hacked that to get it like ULC with the early v9. Tesla just didn't edit it out. I believe he suspects 2019.9 to have additional "goodies" so lets hope he gets his hands on it and his AP unit magically fixes itself.
That being said, I agree with @diplomat33 -- its a great start and portends them delivering by year end hopefully. Who knows though. Getting to 99.9999999% is the hard part (tail end as Waymo called it I think).
That being said, I agree with @diplomat33 -- its a great start and portends them delivering by year end hopefully. Who knows though. Getting to 99.9999999% is the hard part (tail end as Waymo called it I think).
I assume the HW3 larger network with more frames as input, some LSTM and with a larger dataset could take this from 80% to 98%. Then they need a few loops of SW2.0 and a map with some statistical models of the lights to get it to 5 nines.
Similar threads
