So FSDb 11 had a rejected takeoff, once again?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD Beta 10.69
- Thread starter Buckminster
- Start date
Yep, correct. But.. we're talking about automotive radar systems, here. These have been put on $RANDOM cars from random manufacturers for at least 15 years now. By this time, the technology has settled. I suspect that there's a little plug-in module that provides information to the driving computer in a Tesla, complete with information on targets and all that, and that plug-in module and the associated RADAR is acquired from a third party. Why build your own when somebody else has done all the work, has the economies of scale, and can sell it to your cheaper than it would cost to engineer, build, and develop it yourself?
Yeah, yeah: The above is how traditional auto companies have always worked, and look at where it's left them: No expertise in the, shall we say, more complex factors of self-driving cars. But Tesla doesn't build their own windshields: They get them from a third-party vendor, just like everybody else.
But, now that I'm thinking about all of this.. The RADAR-based cruise control systems have been around for a time and have been used on everything from Toyotas to BMWs. And are probably optimized for plug-and-play for that purpose. Suppose that Tesla has realized that they need More than what the standard box can provide, at least for FSD-b. So, they would have (probably more than) two choices: Roll their own RADAR with the advanced feature set they'd like to see or... use their existing expertise with vision systems, for which they've already got the hardware installed, improve the software (which is purportedly getting better by leaps and bounds), and ditch the RADAR entirely?
That.. actually makes a bit of sense. Everybody's been screaming about how Tesla has ditched RADAR as a way of reducing the cost of goods. But we kind of know now that Tesla makes a heck of a profit margin per car, so the couple hundred bucks (if that) may not have been such a big factor. Getting the advanced ADAS for FSD-b might have actually been the more important bit and, if one is going to ditch RADAR and go for vision anyway, why waste the money on software support and hardware in R&D and production and just ditch the RADAR entirely? It's not like one has to redesign the car to leave a sensor connector unplugged.
So, along with all the other new stuff coming down the pike with FSD-b, we'll have Full Vision being made with cutting-edge technology. Cutting-edge, in this case, is a stand-in for Research. Research being defined as, "Running down alleys to see if they're blind."
My take: The vision-based cruise control has some bugs in it that'll probably be solved, right along with all the other bugs flitting about in the code base. But the R&D development engineers have my full sympathy.
L3 is certainly not my only concern. Why not implement L3 initially followed by city streets then robotaxi?
pilotSteve
Active Member
old pilot
Member
So I left my house, turned immediately right and approached a T-intersection I've taken hundreds of times on FSD where FSD drives right up to the creep line stops and then quickly turns left or right. Much to my surprise the car slowed down much earlier than usual and crept to the intersection. What is going on I'm thinking. Then I noticed the town installed a brand new Stop sign yesterday. Damm.
Another intersection ruined by a Stop sign adding 10 seconds to my turn.
Another intersection ruined by a Stop sign adding 10 seconds to my turn.
Pdubs
Ingesting large quantities of LSD
KArnold
Active Member
Financial liability. Tesla has no interest in having an L3 system. No reason to do so.
Of course they do. The take rate for FSD today is maybe 15-20% for FSD. If people could use L3 on restricted access highways and text, read and watch videos on the 15" screen etc. the take rate would be multiples higher. That is a lot more revenue.
I'm not saying you stop there but Tesla is fairly close to L3 now on highways. Driverless FSD (i.e. robotaxi is a long way off). Assuming, that Mercedes does improve their L3 on the highways they could push Tesla to respond. Think FSD revenue.
sleepydoc
Well-Known Member
What you say may be true but even if it is they have no excuse for removing a working system before they have an equivalent system to replace it.
sadly, history shows that is standard operating procedure at Tesla. They have a constant cycle, throw the working system away then regress everyone to a hopeless new version and slowly improve it before giving up and regressing again.
There's a guy on youtube who drives FSDb through Manhattan, very calm, disengages/intervenes when he needs to...but I was surprised at how the software handled the most ridiculously marked streets, with bike lanes, bus lanes, cars and trucks blocking any lane, etc.
Crazy, fun and interesting to watch especially after all the FSDb videos we're used to...enjoy.
Crazy, fun and interesting to watch especially after all the FSDb videos we're used to...enjoy.
It was not a random turn. In the video, the car turned onto the main road and was in the rightmost lane. That lane became a right turn only lane, so the car turned right, as it should have. Apparently, since the car had no destination and there was no slower vehicle to pass, it had no reason to move out of the lane to continue straight.
Also from that Dirty Tesla video towards the end when going across half of a 55mph highway, it seems like there's a "blind spot" in 10.69.25.2 when vehicles move between the pillar camera and main camera:
This closest white vehicle to the front-right isn't visualized at all with FSD Beta drawing a dark blue line indicating it wants to go. The vehicle should have been visible from the wide fisheye camera, so odd that the unified 360º prediction would lose track or drop it under some threshold.
This closest white vehicle to the front-right isn't visualized at all with FSD Beta drawing a dark blue line indicating it wants to go. The vehicle should have been visible from the wide fisheye camera, so odd that the unified 360º prediction would lose track or drop it under some threshold.
Daniel in SD
(supervised)
Looks like it’s just the a bug in the visualizations. The cars disappear for a split second when they change from blue to white.
AlanSubie4Life
Efficiency Obsessed Member
Doesn’t explain the briefly dark blue line with the little lurch though. I suspect you are right and the planner is just randomly wrong as per usual though.
The NNs will fix all this though; can’t get rid of C++ and this archaic procedural code soon enough. It’ll be
. We will no longer be constrained to just the most obvious solutions to vehicle trajectory.
Last edited:
He was mentioning how it darted out when there clearly wasn't a safe gap. Unfortunately it happens too often.
It happened to me this morning. It was an unprotected right turn with cross traffic and a 40mph speed limit. I looked left and couldn't see if traffic was coming from the driver's seat so I know the B pillar couldn't see anything either. But it didn't phase FSDb at - it took the college try. It most certainly was not safer than a human let alone a delusional 300x.
It happened to me this morning. It was an unprotected right turn with cross traffic and a 40mph speed limit. I looked left and couldn't see if traffic was coming from the driver's seat so I know the B pillar couldn't see anything either. But it didn't phase FSDb at - it took the college try. It most certainly was not safer than a human let alone a delusional 300x.
Similar threads
- Article
