-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD / AP Rewrite - turning the corner?
- Thread starter MrBadger
- Start date
In which case, the reference to the EU was more valid, as the USA in general seems to be adopting FSD more readily than the rest of the UNECE.
The other issue Elon has is trying to keep up with Mobileye after dropping them - they're already making claims for L4/L5 and I can easily see major auto-makers offering that towards the end of this year if they can also get approval.
I posted a vid link before of their Munich trial - impressive and a neat uncluttered UI.
I posted a vid link before of their Munich trial - impressive and a neat uncluttered UI.
It LOOKS great. I posted that when I first saw the vid. It was interesting to see that the visualisation of the driveable space is very stable especially for most significant objects. Vehicles that are tagged as immediately significant are also stable. More distant objects have similar perspective / tracking glitchy movement as in Tesla display, but makes sense to not waste processing energy on stuff that is deemed insignificant.
Over all, it already looks like a very realistic (as in appropriate) level of visualisation presented to the driver. Not over cluttered but enough to give confidence.
The other point that struck me (after so much debate over mapping technologies) is the rationale behind fleet updating of map data. It makes a lot of sense to me that vehicles capable of 'seeing' the driving environment should be able to collaborate about how to best navigate it. Whether you want to call it hi-res, ultra high-res or even if Tesla has it or not, the concept seems valid and builds in adaptability and ability for vehicles to modify behaviour in a collaborative manner.
The stability of the visualisation image combined with apparently identifying important objects and focussing on them seems far more evident than in current Tesla implementation.
Some of that stability is achieved by the lower refresh rate of the Mobileye visualisation. I think that video shows that you don't need to maintain a fast refresh rate to have a very pertinent display. I'll take stability over refresh rate any day.
I think though that Tesla may have a tougher job to achieve that with road layouts. Mobileye have the benefit of their higher detailed mapping providing a stable view of the road layout. Tesla seem to be building their road layouts predominantly using vision and AI predictions (much as a human would do without a map) so, as with trying to guess a vehicle, the road layout will change over time as the AI gets more clues to refine it - I see Mobileye have much the same vehicle type, position and orientation changes as the Tesla UI does. This can be seen in the FSD beta 'vector' wireframe view. I'm not sure if/how this will transition to shaded space like Mobileye uses and which I very much prefer.
Wol747
Active Member
Yes, much more stable and visually informative than the Tesla version. And from the small activity on the steering wheel not nearly as twitchy and timid as the FSD beta.
As regards the constant updating of the mapping - which is an obvious requirement - one does wonder what happens to the FIRST vehicle that discovers and reports the workmens' trench across the road?
As regards the constant updating of the mapping - which is an obvious requirement - one does wonder what happens to the FIRST vehicle that discovers and reports the workmens' trench across the road?
As I understand it the ethos of mobileye is that any of their sensing subsystems should be capable of running the show but all should have redundancy. Our Technology
Therefore the first car meeting a new challenge should be able to cope - albeit with hesitancy - but subsequent vehicles will have improved mapping to progressively cope better.
Think new traffic light system/junction on your route - the first time you meet it you may well blunder into the wrong lane too..
Similar threads
