Still it is Sunday and Chuck needs to do his "test" on a workday in the morning to make it "official".
No way will Alan except a Sunday drive as proof.
Enough traffic to be interesting. It's unfortunate it was unable to hit the mark regardless. I remain undefeated. (8/9)
I would argue that #3 was also a failure - the oncoming driver shifted position in their lane to be sure they would clear the back of Chuck's car. (This is a clear failure per the reasonable and permissive rules.)
Agreed as
@kabin says that #6 was also a failure - clearly the minivan slowed down and was not going that much slower than traffic in the middle lane prior to encountering Model Y ASS.
It looks to me like it is working better than 12.3.x -
the main improvement is that it is not stopping in the middle of the near traffic lanes for traffic coming from the right. Those following this closely will recall that was a key critical flaw of 12.3.x.
There's still the issue of stopping miles behind the stop line. Bizarre that that is not fixed, but we saw it in the testing as well, so not a surprise that 12.4.x is also so afflicted.
but it looked like it would have been fine, and even at the end he questioned if that disengagement was appropriate or not.
This would not have been fine - it would have required the oncoming vehicle to take action to allow extra clearance. It takes four complete seconds for the Tesla to clear traffic lanes when there is traffic from the right (from commit to clear). You can see the commit from the torque as Chuck said.
The commit at 2:53
Clear of the traffic lanes at 2:57
The oncoming vehicle was four seconds away when the commit occurred.
The commit at 4:07 (plus a few frames)
Vehicle arrives at 4:11
Definitely a completely legit and advisable disengagement. I have no issues with it.
Interesting that in at least one case Chuck said that there wasn't enough room,
Definitely Chuck is attuned to the extremely slow crossing speed of prior iterations. There is often plenty of time here!
Still, I would say there is hope for 12.5.x, or possibly (likely if traffic is light?) even a random iteration of 12.4.x using exactly the same training, through luck.
I have serious doubts about the ability of it to be successful in more challenging traffic - and we did not get to see it deal with a left-turning vehicle here to check how cognizant it is of what other vehicle plans are.
Not sure whether Chuck will give it another shot or not. Probably not - I would suspect another iteration of software before he bothers with it.
Tesla is really quite close (to the first 9).
- Need to fix the stop sign problems. Why stopping so far back? This is not legal, as has been established. People have received tickets for this.
- Need to be a lot faster after the commit. Can roll it in gradually to allow driver time to intervene on poor commits. Use the available power!
- I think they should just wait for traffic to clear in both directions most of the time; would allow for faster crossing speed and would translate better to situations without a median (median makes it easier!). Problematic in very heavy traffic but for this situation it should have just been rolling most of the time. Personal preference here.
We await the first 9. Then the remaining 9s will fall like dominoes.
