-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla Radar Speculation
- Thread starter Takaa
- Start date
Since Mercedes has a completely different system than Tesla, it would be a very long stretch to assume that it has the same problems as AP.
If MB geofences to high density areas, it may be that they are simply prioritizing their efforts to areas with high likelihoods of traffic jams that fit their L3 ODD.
More potential Arbe Rumors... Japan just certified the production ready Arbe chipsets for use/testing/sale in the Japanese market, and confirmed Arbe is working with a major Japanese auto maker of a sensor fusion system...
My money is on the next gen Honda Sensing 360. 1 front camera, 1 front long range radar (Phoenix) and 4x corner mid range radar (Lynx)
My money is on the next gen Honda Sensing 360. 1 front camera, 1 front long range radar (Phoenix) and 4x corner mid range radar (Lynx)
Here read this and try again on why PB happens and why Mercedes is only certified L3 in high density areas.
Your comment implies that I assert Tesla has no PB issues, which would be erroneous. But, perhaps you could explain the connection between Tesla Vision PB issues and Mercedes multi-sensor L3 system. I just can't find the connection other than both cars have some level of self-driving.
Whether that makes me blindfolded, I cannot say. But, since I do not own a Mercedes, let alone one with an L3 system, I guess it makes no difference either way.
If it's important to you to believe that Mercedes has issues similar to Tesla, that is, of course, up to you, for whatever good it does you.
I thought that it was more that Mercede's system was limited to 37MPH on divided and limited access roads (which basically means traffic jams) but that speed will improve over time such that it can be used in more scenarios (including low density roads)?
Almost by definition, the present Mercedes highway traffic jam system can successfully operate only in high density areas because those are the areas that might have traffic jams on the highways.
Correct. The Mercedes L3 system in its present form appears to be extremely limited (as @ddeflyer just posted while I was writing).
l the Mercedes pioneering L3 system is really very constraied:
- It works only with a lead car
- it works only on high-speed limited access highways
- Yet only at abnormally slow speeds on the high speed road
- It works only with favorable sun elevation ( Tesla has some issues with this also, but the MB system appears to give up much more easily than Tesla's AP)
- It disengages in the presence of significant highway curvature or bank angle
- On a related side note, the MB system has a surprisingly short handoff time upon disengagement, and no ability two get itself out of traffic if the driver doesn't take over properly and quickly. This makes it very marginal as an L3 system IMO. The driver cannot "check out" unless the slow highway traffic jam will be continuous for minutes at a time, and better not be slow to re-engage. Don't wake up groggy or try to finish the current thought in the email you're composing.
- Volvo, for a contrasting example, has described their plan for a highway L3 system, a software enhancement supposedly coming to their new Lidar-equipped EX90. Their idea of L3 is that the car will drive itself at highway speeds without the follow-the-leader traffic jam issues, and normally will give a few minutes of advance warning (upcoming exit or cloud- informed highway hazard condition) for the driver to wake up and get ready to take over. This is much closer to what one might imagine Tesla would do if they decided to offer an L3 version of NoA (not trying to start any rumor here, as there's no evidence Tesla is planning that).
And yes, the systems are so different that I think using its limitations to explain Tesla's problems is probably too far out on a limb. Maybe only in the most general sense, that any AV system will inevitably have some occurrence of false positives - kind of the same as any human defensive driver.
LOL. My comment implies that Self Driving relies upon external reference points in the environment outside. Both Mercedes and Tesla evidently have issues when attempting to find "reference" points in bleak low/zero density areas. My cars have had PB when I was driving on 2 occassions
1. a straight highway with literally no traffic
2. while going up a high elevation bridge at 70mph that resembles a Six Flag roller coaster ride
See picture below
If the Mercedes system is still fusing traditional automotive radar (IDK but likely), then it has the problem that Tesla (and everyone else) also had when radar-equipped: namely, that stationary obstacles in the road become a essentially invisible to the system, because with the poor angular resolution, it cannot distinguish a stationary object dead ahead from a road sign, bridge abutment or whatever that is just outside the. Therefore all zero or very low mph signals are filtered out.
As this is unacceptable for for a reliable AV system, there must be an alternative. The main choices today are cameras, Lidar and higher-resolution radar. Or a fusion of any combination. Of these three, affordable and reliable Lidar is presently the least developed and most costly - a situation that may change in the near future but not for 2023 available cars. Affordable HD Automotive radar is somewhat further along and seems to be coming to Tesla. Camera "Vidar" is affordable and hardware-ready, and nearing satisfactory software development by Tesla, presumably Mobileye and perhaps Waymo, Cruise at. al.
But it's not done, and again all of the above will suffer from some degree of false positives. I don't dispute your "losing a reference point" comments in some situations, but that's a very general description. For people, it could apply to pitch darkness, snow blind blizzard and smoke camouflage in battle. Three very different things that confuse our senses.
The first certainly relates to your "losing a reference point" comment. The second, I think, is an important bug relating to the use of map data that is imperfect or not sorted properly for overlapping roads. I recently posted a problem I have with a single level road where there is a median U-turn control signal, always causing FSDb to try to stop in the through lane despite it having no relevant signal. This is a similar map issue.
This was not a map issue. I was on simple AP. not FSDb. The car started going up the ramp at 70mph. No issues but as it went up, it suddenly had a "panic attack" and slowed down to 50mph. There are no signs that even stated the lower speed too, like often found on exit ramps. It was just like a roller coaster when you are at the top. Nothing ahead of you and the turn to the left or right at the top is invisible unless you are right there.
I've driven over most of the DFW major interchanges without PB events, however, perhaps you found one where the car loses sight of the road altogether. Drive up US287 to Amarillo and you'll find a few cases where the hills cause the road to fall outside of the FOV of the cameras right as the car crests the hill. Several versions ago, this would cause the car to throw the Red Hands of Death and NOA would bail completely. It was quite repeatable.
Perhaps I should clarify in the Volvo case, that they do not describe it as an L3 system. Volvo is clear that they don't want to use the SAE levels in their public communications, they want to describe it as Supervised=Volvo-responsible versus Unsupervised=driver-respnsible. Of course, eschewing the L3 description also means that they can't claim "first to L3" as MB is anxious to do - but to me it's clear which one would be useful and which one is more of a gimmick.
Similar threads
