odysseus2000 wrote:Tesla auto pilot in uk city driving: Impressive but still needing some manual human interventions. Can it be developed to go to full no steering wheel driving?
https://youtu.be/sk0eZRVw9x4Regards,
Thanks for the link. Very informative video. But also sadly rather disappointing.
When you hear stories about Tesla's supposedly being pulled over by the police with a driver completely out of it drunk, yet the car didn't crash (
https://www.afr.com/news/world/north-am ... 203-h18mnh ) (*), and so on, you get the impression that Tesla's autopilot is almost autonomous driving, only without the final nod of approval to call it such.
But that video completely obliterates that perception in my mind.
I think the driver is perhaps being a bit too generous - I presume the car is his own, purchased with his own money (?) so perhaps there's an element of post purchase justification.
For me, the whole plus of self driving is being able to hand over to the car to do the full commute in exactly conditions like that ... slow moving traffic, crawling from traffic light to traffic light. Who wouldn't want to be sitting back and reading a book (or whatever) and letting the car deal with the traffic?
But for Tesla. they clearly still have a long way to go.
At one point when a car pulled in front of him from the left lane, the driver had to take over and admitted he felt that if he hadn't it probably would have crashed.
Most worrying of all, is that it didn't spot the cyclist (@15.30 he was clearly there in reality and easily visible). The driver did try to convince us he thought the car reacted, but a little too late. I'm not even convinced by that. The cyclist was simply not being shown in the 3d surroundings model. The car should have been flagging the cyclist as present and a high risk a reasonable distance prior - well in enough time to take appropriate action - the highway code specifically says that you must overtake cyclists with plenty of room, incase they need to swerve for a pothole or similar. That Tesla didn't even spot the cyclist, let alone manouvre in a manner that followed the highway code advice.
Nor did the car seem to properly spot the people crossing @11.00minutes in. The driver seemed pleased it fleetingly spotted the person crossing. That's unacceptable ... it needs to be solidly and stably detecting a person in that situation.
Similarly at 10.43, the model shows one single person stood at the side of the road, and bizarrely one stood right in front of the car in front of him that's moving. Yet in reality you can see at least 10 people, all of whom are in a position where they wish to cross the road and at a high risk of stepping out.
And that is simply unacceptable for any car that even pretends to be autonomous.
It's exactly those situations (people and cyclists) why I believe that lidar (or equivalent) will have to be a regulatory 'must'.
That Tesla was quite simply not identifying, and therefore not taking properly into account some of the biggest risks. In those situations (town centres) pedestrians and cyclists are by far the most important things as a driver you need to be watching out for. And that car simply didn't even appear to be seeing that they were even there.
In contrast...
Now I'll fully admit I've not seen real world - independent - evidence uncensored by google / waymo, but here's a video posted by Waymo themselves (so obviously not impartial, and the video will have been carefully selected - I acknowledge that)...
https://www.youtube.com/watch?time_cont ... u8gmFhiGko That video shows the system much more confidently recognising each individual person. And most importantly, keeping a stable representation of them in their representation of the world around the car, that indicates it wasn't just a fleeting 'maybe' that seemed to be quite common in the Tesla video ... although in most cases, I don't think that Tesla video 3d representation really seemed to show all that much as being 'stably' recognised at all!!...
... when waiting at the lights, the Tesla couldn't even stably see
the car right in front of it! ... i.e. between the lorry on the left, and the lorry in front of the car in front - the car a few feet in front wasn't even being shown briefly @ 8.32 into the video! And then it showed it drift left completely into the lorry, while in reality both the car and lorry were both stationary!! That's a blooming great car, not even a small cyclist, and it's only a few feet in front of it and it can't even recognise it as being definitely there! That's completely unacceptable! ...
... If Musk wants to convince us that Lidar isn't needed, then his autopilot needs to do a heck of a lot better than that!Here's a video of a Waymo car not just positively and clearly identifying a cyclist, but also the 3d model already showing a prediction of the expected cyclist's behaviour, and the car giving the cyclist room for the cyclist to perform the expected manoeuvre. That seems to be light years ahead of what we're seeing in that Tesla video where it didn't even recognise the cyclist in the first place!...
https://www.youtube.com/watch?v=NG_O4RyQqGEIn the Tesla video, the driver had to take over fairly often in just a few miles.
Here's the status for Waymo (though again, admittedly this won't all be on city streets)...
https://www.digitaltrends.com/cars/waym ... oad-users/ "Waymo not only led all other companies in total miles driven in the state but also turned in the best performance, as measured by disengagements — instances when the autonomous system required the human autonomous vehicle test driver to take immediate manual control. The Waymo driver went 11,017 miles between disengagements while the second place performer, GM’s Cruise technology, reported disengagement every 5,205 miles. Most companies in the California report traveled much shorter distances between times when the human had to take over. Apple’s self-driving testers, for example, disengaged every 1.1 miles."
Here's another example of where Waymo's at ...
https://www.theverge.com/2018/12/5/1812 ... t-cost-app If you compare where Waymo appears to be at, compared to the experience of that Tesla driver, I know which my money would be on.
Watching that Tesla video (and comparing it where Waymo appears to be at), to me, Tesla seems like a wannabe... substantially way behind the front runners in terms of autonomous capability, but trying to rush it out to users to try to give the impression they're something that they really aren't.
Does anyone watching that video seriously think Tesla is anywhere near matching the 11,000 miles between autonomous disengagements that Waymo is achieving? They are world's apart! (even making some allowance for weather, traffic, roads, etc).
Seeing that video has genuinely set back my estimation as to where Tesla cars are right now in terms of autonomous capability. And not just by a small amount either.
Personally, seeing that video ... I think Tesla are going to struggle. They aren't going to meet expectations in the next 18months or so. Musk is going to start feeling some real pressure.
And over the next 18months or so, there's every reason to believe that Waymo could quite easily, and quietly leap frog ahead.
I really do believe the slow and measured, safety first approach, coupled with huge investment that Waymo has undertaken, is likely to see them really speed past all others.
Waymo has now got a strong technical base to leap frog their business forwards. They've put huge investment into R&D that gives them a real strong platform from which to grow.
I think Tesla is at real risk of 'doing an Uber' ... Uber (in my view) also tried to rush their technology out quickly, didn't test it properly, didn't take safety as seriously as google, didn't recognise the need to do it right first time... and look what happened to them... they ended up having a serious accident (fatality) that significantly knocked back their efforts.
I think Tesla / Musk have tried to jump too soon, from a platform that doesn't yet have solid foundations, and from a height that is just too low to afford them any real chance.
Very interesting and informative video of the Tesla though.
---
(*) Though searching for that story again, did throw up this ...
https://electrek.co/2018/08/25/tesla-mo ... uck-drunk/
"I have to say “yet another fire truck” because it is the third accident involving Tesla vehicles reportedly on Autopilot and fire trucks this year alone"