Yes,
I agree, but at least now It does break for stopped cars.
Tesla Russian Roulette Suicide Mode seems to be a thing of the past.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Phantom braking, etc. overflow from 2018.32.2 thread
- Thread starter dailo
- Start date
I agree, but at least now It does break for stopped cars.
Tesla Russian Roulette Suicide Mode seems to be a thing of the past.
It's better at doing so but I have still seen some cars where AP2 reacts much too late for, at 60mph. At 30mph or so, it seems pretty reliable…. But I still need to intervene maybe 5-10% of the time at 60mph.
I genuinely feel more confident that it will stop (eventually) and rarely if ever intervene when driving alone. When their are passengers in the car, I will often intervene just to prevent a hard breaking incident.
The last time I had an opportunity to try was 2018.26, AP2, 60mph approach to a Prius. The adjacent lane was open and I had to emergency brake as well as swerve. No FCW or AEB kicked in and I came to a stop past the Prius's rear wheel well.
Maybe you can argue that if I just braked in place instead of swerving, my stopping distance would've been shorter. But that was far closer to an impact than I wanted.
Note that most other times, indeed, the car just does somewhat hard braking and comes to a stop normally. But it's still not 100%.
I've never tested beyond 50mph and not recently but I've found that about 10% of the time it would fail to react sufficiently for me at 50mph. I never did @chillaban had the stones to do because I love my car too much. 
I wouldn't fully trust AP. I wouldn't even be confident at this stage. Elon himself says drivers that trust AP often end up in those much reported accidents.
I wouldn't fully trust AP. I wouldn't even be confident at this stage. Elon himself says drivers that trust AP often end up in those much reported accidents.
Yeah to be clear: I don’t recommend doing what I did. For the rest of that drive home I felt like a moron.I've never tested beyond 50mph and not recently but I've found that about 10% of the time it would fail to react sufficiently for me at 50mph. I never did @chillaban had the stones to do because I love my car too much.
I wouldn't fully trust AP. I wouldn't even be confident at this stage. Elon himself says drivers that trust AP often end up in those much reported accidents.
Someone's gotta do the exploration and report back. I appreciate it and aren't judging because I do lots of pushing boundaries but nothing about my experience at 50mph encouraged me to go further.
I said I am more confident but not 100% confident. Most of my Russian roulette suicide mode experiences were in Palm Desert, California where Speed limits on surface streets with controlled intersections are often 55 mph as was the case in the video I posted. Haven’t tried it at those speeds in some time. Perhaps speed is the big factor and not because improvement‘s in software...
I really think it’s both. I think .21.9 and then .24 were both improvements in stationary braking performance. Right now below 30mph (as long as there’s no curves leading up to the stopped car) I’m basically not worried at all. It’s only above 50mph that I get really paranoid.
FYI tonight on .28.5, I was approaching a stopped Ford Explorer around a curve at 40mph.... AP2 made a very sudden last-minute brake for it... so there’s no magic fix in .28.5 either.
However, overall 28.5 seems more smooth and incrementally better AP performance compared to .26.
Yes, this has happened several times, the latest being Aug 22 when the car suddenly applied the brakes with
this has happened to me quite frequently as well, the latest being on August 22 when the car suddenly applied the brakes with the appropriate warning sounds because of a car in a left turn exit lane in front of me. There were no cars directly in front of me in my lane which was quite surprising that the car could not differentiate that. When that happened I was wondering if there is a way to communicate these instances directly to Tesla because I could pinpoint the exact time that happened which should be in my car log someplace.
I thought so too, up until Thursday past when driving my new S100D (AP2.5 hardware, 2018.28.5 software) on the motorway in Autopilot with follow distance set at 1, doing 100kmh (62mph) in the fast lane on a sweeping left bend when a leading car moved aside and mine started to accelerate to the set max of 125kmh (78mph) just as the view opened up to reveal completely stationary vehicles in both lanes about 200m straight ahead. It kept accelerating well past the point where it should have started regen already and I then brought it to a stop with about 75% of a full emergency braking about 10m from the jam.
Another hairy situation encountered on the previous software (v.2018.26.1) a few weeks previously was at a construction site on a motorway exit ramp, where the 2-lane traffic has been shifted a half lane to the left. There were temporary yellow lines marking the modified and narrower lanes. I was in the slow lane at about 70kph following a bus which for no apparent reason shifted into the faster lane just as we crossed the exit. At this point, with no vehicle ahead in the straight lane, my S100D started to accelerate to the AP set speed of 100kph while the steering tracked suddenly about 20° to the right, aiming straight for the end of the temporary steel crash barrier between the motorway and the exit ramp, which was surrounded by a few traffic cones. I quickly jerked the wheel left, back into the lane.
I then decided to turn around and try to recreate this scenario at the same place whilst filming, but did not succeed, as this time the car sailed through with no problems on AP, possibly because the second time around no large vehicle was just ahead and switching out of his lane at precisely the wrong moment to present the system with a confusing situation in which it somehow made the worst choice, highly reminiscent of the tragic Walter Huang scenario.
IMHO in both these cases where the system must (one would hope!) have some doubts about what it is seeing, then it should err on the side of caution and at the very least not ramp up the speed to the set limit into the situation. Even better would be to rather warn the driver to immediately resume full control while gently decelerating. FCW and AEB were active in both cases, but to no noticeable effect.
All in all these are the real wake-up calls which effectively remind us to never rely on this AP thing at all, but keep eyes on the road situation just as much as when driving without it.
Here’s hoping that v.9 will soon bring huge improvements in AP safety/reliability, but I have no confidence that it will in fact do so. If occasionally accelerating into massive stationary objects on the motorway has not been fixed by this stage, then I severely doubt that it ever can be remedied with the current sensor suite and/or programming team. Hence Tesla would be prudent to upgrade both ASAP, prior to being hit with its first contributory negligence lawsuit for the death of an innocent third party.
Similar threads
