-
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 will get a lot worse before it gets better
- Thread starter conman
- Start date
-
- Tags
- autopiliot autosteer braking
Thank you for your perspective. I do agree that false-positives are unpleasant. However, I think I take them much less seriously than you do. To me they are just annoyances that are clearly being worked on and improved with every firmware release. When I first got my AP2 it was extremely iffy and I would say dangerous. The car swerved often and there was a lot of phantom braking. I still loved it, the frequent imperfection being like a challenging video game. I watch the road very intently when on AP, ready to respond to any improper moves. In the last two years AP has improved tremendously, to the point when I seldom have to make any corrections. It is so much improved!
I am also wondering if it depends on where you live and drive. I often hear about the following distance being too long even when set on 1. This is extremely perplexing, since I have it set at 7 and often wish it was longer. I wonder if in certain locations people drive much more aggressively and leave a lot less distance around their car for the occasional phantom braking to be an issue. Same goes for "long" following distance -- I really cannot imagine leaving any less distance than "7" in traffic, while others have commented that other drivers get annoyed and cut them off when set at "1". AP may not be as usable in places like that.
I own two Teslas, and I am sometimes embarrassed to own such expensive cars, but the expense and related embarrassment is not why I own them. We own these cars because they're the best BEVs you can currently buy, and to us that is worth the embarrassment of having spent so much money on them. What about this is contradictory?
Yes it does. I live and commute in an urban area with dense traffic and people driving like maniacs. The highways have tight turns, tight entrance/exit ramps, narrow lanes, often no shoulder, and many bridges/overpasses. TACC works best in this environment when taffic is stop & go. It does poorly when traffic is flowing >45mph but is highly dense. Autosteer gets tripped up by many things in this environment even when traffic is light, so I often use only TACC.
Thanks. I suspect more and more people will agree with my post as FSD features are rolled out, as I predicted...Well, this turned out to be a prescient thread based on my current experiences
I concur and relate w/ your experience. While recognizing that the Driver Assist features in my Model 3 are far from complete, I am always ready to intervene if need be. I am actively monitoring what the car is doing (which, by the way, is exactly how autopilot works in planes - it MUST be actively monitored at all times) - and find that over the 6 months I’ve owned the vehicle, I’ve had to intervene less due to software updates. This is w/ daily driving in upstate New York and several long road trips. I realize that YMMV, but I’ve seen mostly improvements to AP and expect it to get better over time, even if occasionally a software update introduces a new bug.
@verygreen can perhaps show a live feed video when it's phantom braking and what it sees.
Phantom braking the later versions has been horrible!!! What exactly is going on? What does it see ?
Phantom braking the later versions has been horrible!!! What exactly is going on? What does it see ?
Considering mine still occasionally misses vehicles in certain settings (especially if they're dark grey/silver and similar colour to the bitumen) when they're completely stationary, they're simply recognising more and more things as vehicles in the neural network. Unfortunately more of the things they're recognising are simply bits of road that look slightly different, especially with change in road material, inclines, declines, shadows at underpasses and so on. Additionally the cars in the other lanes all look like they're about to change into your lane, according to the neural network, so it's slowing down for them.
I had a couple today and the car did not see anything (or at least I am not seeing anything strange).
It clearly was spooked by oncoming traffic in my case, but those cars are not marked as obstacles or anything like that.
Thanks! That's what I kinda suspected. So it has to be some kind of a POSE problem or some other bug in the code?
Or it can be the radar code that's getting worse for some reason. Who knows
What has changed over time is the weighting of the information from the radar versus the camera. Due to the recent accidents there is more weight on the radar now than before.
Now that HW3 is the new hotness, I really fear for where they will call HW2.x performance good enough and quit improving it. I mean, they never made any guarantees about phantom braking, right? They can just quit improving HW2 at any time and focus all efforts on HW3. I could argue that having purchased EAP (the no-longer-available option), I am entitled to "on-ramp to off-ramp" which NoA isn't really yet, and also I'm entitled to some kind of park seek and advanced summon, which haven't been delivered. But Tesla can be very forgetful about promises, we know. And anyway, they can deliver unusable, toy versions of all of those things with unbearable phantom braking, and still realize the revenue and move on, if they choose to.
golfpilot
Member
I don't have any badges on mine. I'm regularly telling people "this is the Tesla you can get for 35k".
Nice. I seem to be missing something. What's the relationship between that and phantom braking?
This was a continuation of a moderately tangential discussion that happened months ago in this thread. I think it started when somebody asked about phantom braking in ACC systems from other manufacturers and I piped up with my impression of the 1st-gen Prius ACC, and then it spun a little off-topic.
There is even a term for this used by the people who know what they're talking about: sensor diversity. Elon may be correct that you don't need lidar, or that "lidar is a crutch", but that misses the point. Sensor diversity, even if not strictly required, makes the system perform better in a variety of circumstances.
I'll go ahead and make a counter-argument against myself though. Every sensing type has false positives. If phantom braking is caused by false positives, then more sensing types means more false positives, and therefore more phantom braking.
Counter-counter-argument: If you use the sensor diversity to seek confirmation of one sensor's detection before trusting it, you can reduce false positives.
Counter-counter-counter argument: If you ignore what one sensor is telling you because another sensor doesn't see anything there, it may be just that the other sensor is having a false negative, and there really is something there, and by ignoring it you're going to get a lawsuit and lots of bad press. (e.g., ignoring radar return from the side of a truck because the camera thinks it looks like sky.)
Counter-counter-counter-counter argument: If you have at least three sensing modalities and sophisticated sensor fusion, it becomes increasingly unlikely to have a false negative on two modalities at the same time. (e.g., Lidar+Radar would see that truck and camera would be outvoted.)
You can keep going back and forth. I leave this as an exercise to the reader.
Alternatively, the more things it will find a reason to brake for...
EVNow
Well-Known Member
The reason Tesla doesn't have LIDAR is that it is not practical in retail cars (well it wasn't at that time, anyway).
So, Tesla started wondering whether AP/FSD might be possible just using vision. Turns out MobilEye thought the same way. So, they partnered.
The rest is history. Its one of those things where necessity has become a virtue.
Similar threads
