-
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 getting worse - any ideas on how to stop it?
- Thread starter Ev_Rider
- Start date
Only time I've seen FSD Beta drop speed on the freeway significantly is when GPS thinks I'm off the freeway and on a side street. This is due to construction in the area that has shifted the lanes. When the lanes shift to the right in the construction zone, the map shows the arrow off the freeway and on a side street (usually an exit lane that has left the freeway but running parallel).
I know that in construction zones, where lanes are shifting, I disengage and drive manually through it.
Phantom braking has been like this since 2016 when Musk decided to ditch MobilEye, and was supposed to do much better himself. My car is a 2018 model, and has had such braking from the first day I got the car. The service center has given up, and now says that there is nothing to do, and that it is "by design".
Braking implies the car THINKS there is something right in front, doesn't mean it is. A mirage can be misinterpreted by the system as being a close obstacle.
Which is a failure of the overall system, of course, and people jump all over that and start talking up LiDAR and RADAR. What isn't discussed is whether a vision system can be trained to reliably filter out heat shimmers and other such visual effects. I think that they can be, but I also think that in practice it's not worth spending the time. That may be why Tesla is going back to RADAR for Hardware 4. They called RADAR a crutch back when, and I think they're fine with using it as such so they can spend their development resources on other stuff.
Honestly, I think Tesla would have gone with LiDAR and RADAR from the start if they were as cheap as cameras, but they needed something that would work for the mass market. They were a little too optimistic about doing everything with inexpensive cameras combined with lots of training. My understand is that the RADAR unit in Hardware 4 is a custom job that was designed by Tesla. That suggests to me that they've seen the writing on the wall for while.
I wonder if Tesla will allow the Hardware 3 RADAR unit to be retrofitted so we can all have the benefit of it.
or at least re-activate the radar from cars which came with radar and HW3 and had then later the radar software deleted...
from reading up on it the "HD Radar" isn't anything fancy... but custom build for Tesla with focus on cheap rather than being especially capable. Apparently it's good for ~100 meters ahead of the car which isn't anything to write home about but telling how little Tesla trusts "vision only" even that close in....
The New High-Definition Radar in Tesla's HW4 Is Not As Revolutionary as Promised
Tesla made a surprising return of the radar sensor with the HW4 sensor suite, but it doesn't seem to be the high-definition radar that they promised
www.autoevolution.com
My RGV to Houston drive resulted in zero PBs on 2023.20.7. It was my first time driving that route tho in the Tesla so not sure if it’s just a route that doesn’t trigger it. The real test will be when I drive to El Paso from the RGV in a few weeks. That drive has always been PB hell. Will update
curious to hear how it went. as of now the wife has vetoed further long road trips in the car until PBs are much much reduced. Taos to Ft. Worth via Amarillo was hell... every 5-10 minutes a hard braking attempt
July 2019 Model 3 AWD with HW3 ... but radar deleted since August 2022 .... zero issues on the same route in all years prior *with* radar enabled
I drove from Santa Fe to OKC yesterday, mostly on I-40 across New Mexico, the Texas Panhandle and Oklahoma. 99 degrees, sunny, with lots and lots of mirages above asphalt and concrete. I was in a MY with FSDb 11.3.6. I had zero PB events along the entire stretch. A week ago, I drove out there on the same route. It was a bit cooler and cloudy much of the time. I also had no PB events, but was guarded as to the effectiveness of the single stack FSDb since the conditions were not typical for summer. Yesterday's trip satisfied that.
Last summer, using the old NOA, I had numerous PB events on I-40, so I can say that Tesla has made great strides in PB reduction. Now, if they would base basic AP off of FSDb, I think that PB will be mostly a thing of the past.
If the road has the right vertical profile, as with short rolling hills, it may be possible for a reflection point to move rapidly toward the car. In theory, if the perception software cannot control for this, it could interpret the reflection as an obstacle on a fast collision course with the car.
A year ago, I was able to see phantom cars caused by road reflections. It was trick to do, but I managed to position my car just right well back from another car that was too far to be displayed in the visualization. If the road reflection point was just right, I could see a the car in the display much closer than it was. However, I was not successful in producing an PB event from it. This might have been because I just didn't get the correct dynamics to get the reflection point to move toward the car at the same time that another car was falsely identified. Or, it's possible that mirages were not the culprit at all.
In my 50 mile experiment between Clayton, NM & Dalhart, TX, the mirages were not present. That is why I am ruling it out. That said, it is also possible that there are more than 1 reasons for PB. I just didn’t experience any due to mirages.
Similar threads
