Elon:
"Already testing traffic lights, stop signs & roundabouts in development software. Your Tesla will soon be able to go from your garage at home to parking at work with no driver input at all."
(tweet)
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Elon: "Already testing traffic lights, stop signs & roundabouts"
- Thread starter strangecosmos
- Start date
Just ship it already. It can't be worse than NoA.
Tam
Well-Known Member
I thought those were supposed to be tested out in 2016 to prepare for scheduled 2017 coast to coast trip.
It's good to know that they have finally tested those out at last.
They clearly did not have the right people on the team back then. That testdrive was mostly hardcoded and not structured.
Karpathy did confirm a complete rewrite released March 2018 (2018.10.2), which seems to lay out the fundamentals required to add the functionality now being rolled out (NOA) and tested (traffic lights etc...). Plus they need to roll out the new AP 3.0 hardware for the more advanced NN to work.
I'm looking brighter at things now than back in 2017.
I have a hunch (which may be wrong) that Tesla is approaching the problem of full self-driving step by step. Create a solution for one modular task like highway lane changes or stop sign detection, get it up the level of the reliability where it's ready for production release, release it, and iterate on it.
Elon mentioned on Twitter that the coast-to-coast demo will probably be done with Software V10 alpha. So rather than being a proof of concept or "demoware", the demo will be done with software that is in the production development pipeline.
Elon mentioned on Twitter that the coast-to-coast demo will probably be done with Software V10 alpha. So rather than being a proof of concept or "demoware", the demo will be done with software that is in the production development pipeline.
Canuck
Well-Known Member
"Already" -- wow that was fast! And here we thought FSD testing wouldn't take place until many months or years from now, especially since he has already showed us a FSD video more than a year ago, and also what he told us more than 2 years ago now...
Fully autonomous driving? Elon Musk promises Tesla’s cars will make it so
... and was selling to us until only recently. I can't believe they are already testing it!!! That's a huge surprise. Thanks Elon for keeping us updated.
(Sarcasm in case anyone missed it.)
Last edited:
It really is a bit unfortunate on a such a small forum as this particular autonomous one is a volume poster like @strangecosmos has so many of the people on his ignore list that he keeps duplicate posting almost every topic already discussed everywhere by the ignored ones.
It is like having someone having a completely separate conversation as the rest of the people in the room.
It is like having someone having a completely separate conversation as the rest of the people in the room.
Last edited:
At the current rate of progress for the past two years I would guess that we will get the ability 'to go from your garage at home to parking at work with no driver input at all' in about 2040 unless they hit any snags. The car still can't even park itself properly even if you have it lined up with the parking space. It is far too slow to be of any use and most of the time it simply gives up before it has finished leaving the car sticking out and requiring manual intervention.
As for driving on the road it can't see potholes or small obstructions. It gets confused at road works and many, many intersections. I can imagine having to go out and rescue my car when it has stopped in the middle of an intersection no knowing where to go.
At the moment it is a neat bit of future technology that is better than the other cars on the market but a very long way from being truly useful or anywhere near reliable.
Tesla will get there but Elon needs to step into the real world and stop promising things like this which are so obviously wildly inaccurate. A year ago we were supposed to see a Tesla drive across the USA. What happened to that. Once that has been done it will still be a couple of years at least before it is ready for us to use. I don't expect my Model S will still be going when it happens.
As for driving on the road it can't see potholes or small obstructions. It gets confused at road works and many, many intersections. I can imagine having to go out and rescue my car when it has stopped in the middle of an intersection no knowing where to go.
At the moment it is a neat bit of future technology that is better than the other cars on the market but a very long way from being truly useful or anywhere near reliable.
Tesla will get there but Elon needs to step into the real world and stop promising things like this which are so obviously wildly inaccurate. A year ago we were supposed to see a Tesla drive across the USA. What happened to that. Once that has been done it will still be a couple of years at least before it is ready for us to use. I don't expect my Model S will still be going when it happens.
Soda Popinski
Member
I don't know about the end of 2019, but 2040 seems a bit on the pessimistic side. You're judging current HW2/2.5 capabilities. The HW3 upgrade uses a dedicated NN processor capable of something like 10 times the computational power of the current GPUs.
Seattle Tom
Member
I don’t know if HW3 will really make it better, but I’ll wait and see. I don’t think anyone can positively know the answer to that yet.
But the tweet from the OP and the idiotic Elon comment from the Q4 call (“we already have FSD on highways!”) makes me think nothing is coming soon.
If Tesla judges the NoA we have as “FSD on the highways,” then the gap between what I think is reasonable and what they think is reasonable is HUGE. It’s a nice Level 2 system - better than what I’ve had in the $30K cars I previously owned - but someone who says that is FSD is not even on the same planet I’m on.
However, the good news is that I’m sure the new CFO, who was an intern at Microsoft 8 years ago, will bring some serious professionalism to the strategic plan and execution. Oh God...
But the tweet from the OP and the idiotic Elon comment from the Q4 call (“we already have FSD on highways!”) makes me think nothing is coming soon.
If Tesla judges the NoA we have as “FSD on the highways,” then the gap between what I think is reasonable and what they think is reasonable is HUGE. It’s a nice Level 2 system - better than what I’ve had in the $30K cars I previously owned - but someone who says that is FSD is not even on the same planet I’m on.
However, the good news is that I’m sure the new CFO, who was an intern at Microsoft 8 years ago, will bring some serious professionalism to the strategic plan and execution. Oh God...
favo
P3D+ owner
I, also, was surprised that Elon proclaimed FSD achieved on highways. However, he's probably used to the latest developer builds, and may have forgotten what problems the released version has. Having said that, I recently drove 800 miles on AP to Atlanta and back, and it was very good overall.
Also, there are plenty of smart people who work/worked at Microsoft (I know a few personally).
It was 2006, according to his LinkedIn page, so your math is a little off.
Also, there are plenty of smart people who work/worked at Microsoft (I know a few personally).Up until it actually has to change roads. NoAP can't even handle half of the Bay Area interchanges, like CA-17/CA-85 or the entrance ramp from the Mountain View Supercharger onto 101. No, it's not even close to full self driving yet.
BTW, I had the opportunity to drive Old Santa Cruz Highway on autopilot today, thanks to CA-17 being closed in both directions. There were enough cars that I averaged about 3 MPH, so I felt comfortable letting my car drive on that horrible road.
The record was, I believe, four disengagements in a single minute.
No, not even close.
I'm both optimistic about HW3, and pessimistic.
I'm optimistic because I strongly believe HW3 will improve the accuracy of the neural networks. In my own testing of NoA I noticed that it would fail on auto-lane changes around 10-20% of the time. It seemed like it was getting thrown off by ghost cars. Cars it thought were there, but weren't there. I also find NoA a bit awful because it doesn't know the speed limit a good portion of the time. HW3 should bring sign recognition.
But, my absolute biggest issue with NoA is the lack of good maps. For example last time I drove down 405S to I5S it didn't know what lane I needed to be in. Those two freeways are two of the most popular freeways in our State. How could they not have that? It also has issues with I5N through Seattle where it tries to get me to take the Seneca street exit.
Tesla needs to have something in the UI to report map issues.
DDotJ
Member
Yes! NoA has difficulty with which lane it needs to be in, especially with HOV lanes. On the 5 Fwy in SoCal it keeps attempting to take the wrong interchange ramp or will not recognize that there is an HOV interchange ramp. So it continuously tries to get me out of the carpool lane to take the non-HOV ramp even though there is a carpool connector (91 Eastbound to 5 SB ramp). When going straight on the 55 Northbound it suddenly tried jumping onto the 5 north HOV ramp causing me to disengage AP and override it.
It is also blissfully unaware of where carpool lanes open up and close, so NoA tries to get me back into an HOV lane 1.5 miles before an exit. In reality, the lane would close and I would miss the exit.
It's also inconsistent in how it want to change lanes. Sometimes it will keep me in the far left lane with less than 1 mile until the exit, forcing me to override and put the car in the exit lane. Sometimes it wants me in the exit lane 4 miles ahead which puts me behind slow trucks and merging traffic.
On the 405, it was miscounting the number of lanes on the left. I was in the leftmost lane next to a concrete barrier and the car showed multiple lanes to my left.
I tried emailing [email protected] and at first they responded saying they would look into it. Now they no longer even respond.
I wish they had something like the Google Maps Maker (editor) or something for correct lane guidance and NoA.
Last edited:
Isn’t Tesla claiming they have a generic solution that does not rely on detailed maps?
