Autonomous Car Progress

[acoste]

So, Waymo has similar phantom braking issue - and also may try to hit pedestrians.

I thought Lidar solved those issues

This I think is not phantom braking (false positive detection) but more like inaccurate prediction of pedestrians'/cyclists' future path.

Surprising that it missed some pedestrians.

 

[electronblue]

I thought Lidar solved those issues

Lidar does pretty much solve false negatives but nothing of course suggests it alone solves false positives especially when a multitude of sensors is concerned.

That said while I do trust customer experiences in things like phantom braking or unnecessary delay I am less inclined to believe the accuracy of reports on missed pedestrians — unless the car actually hit them. A passenger is probably not really equipped to know if the car saw it or not given that pedestrians also can have unpredictable paths and the car may choose to ”ignore” them up to a point (until they get on the path of the car) for obvious reasons.

Then again even Uber saw the pedestrian they killed — they just didn’t listen to their Lidar.

Complex issues. Luckily Tesla is leading the way with Level 5 no geofence feature complete by end of 2019!

 

[EVNow]

Lidar does pretty much solve false negatives but nothing of course suggests it alone solves false positives especially when a multitude of sensors is concerned.

That said while I do trust customer experiences in things like phantom braking or unnecessary delay I am less inclined to believe the accuracy of reports on missed pedestrians — unless the car actually hit them. A passenger is probably not really equipped to know if the car saw it or not given that pedestrians also can have unpredictable paths and the car may choose to ”ignore” them up to a point (until they get on the path of the car) for obvious reasons.

Then again even Uber saw the pedestrian they killed — they just didn’t listen to their Lidar.

Complex issues. Luckily Tesla is leading the way with Level 5 no geofence feature complete by end of 2019!

This clearly shows why people have difficulty believing in things that disprove their theories.

You first statement is contradicted by the report that the car almost hit pedestrians. So, you don't want to believe it.

 

[emmz0r]

Thanks for the vid. there are some interesting points.

At the Y crossing for one moment it thinks the gore zone is drivable and it also offers the possibility of changing lanes into the gore. Just like how Tesla accidents happen.

View attachment 446853


View attachment 446854





One good thing! As soon as the car on the right touches the divider lane (switches lane into ours), it is labeled as "in our lane". Tesla isn't very good at localizing lane changing cars in front.

View attachment 446855

Doesn't the nVIDIA car have a christmas tree of sensors on top as well?

 

[electronblue]

This clearly shows why people have difficulty believing in things that disprove their theories.

You first statement is contradicted by the report that the car almost hit pedestrians. So, you don't want to believe it.

Oh, I want to believe whatever is accurate. I have nothing vested in anything but Tesla (an AP2 car with FSD). But I’ve dealt enough with consumers to know things are not always quite what they seem.

When it comes to missing or hitting pedestrians, I did not walk away from the report with the belief that the consumer reports necessarily were accurate on that account. I would have to know more about the incident to form my own opinion on that. Maybe they were, maybe they weren’t, just offering a point.

Someone said it was surprising Waymo misses pedestrians. I said I feel we’d have to know more. If it misses clearly visible pedestrians crossing a street straight ahead, that would be very different from a pedestrian ”appearing out of nowhere” or changing direction unexpectedly etc...

Also, as said, the customers may not really know what the car saw and what/why it decided on such things.

There are things where the report is more fruitful and where customer reactions provide more immediately useful insight: Waymo’s hesitation, route selection issues and so forth are clearly demonstrated and those I believe. But missing pedestrians is a bit different in the sense that they can be unpredictable and a customer may have limited visibility in how the car saw or judged them.

If you have more details on the ”almost missed pedestrian” that would make me a believer, feel free to elaborate.

 

[Tam]

...the customers may not really know what the car saw and what/why it decided on such things...

It's certainly nice to know how it happened but I still think what counts is a layperson's experience.

It's like when a layperson saw the returned rocket did not land on the barge but it actually fell into the sea instead.

A scientist might rationalize that it did not actually "fell into the sea" because according to the simulation and all the calculations, it should be rested on the barge by now and not on the bottom of the sea and what happened was one of the engine nozzles was just a little bit short of fuel, that's all. No big deal!

 

[acoste]

Oh, I want to believe whatever is accurate. I have nothing vested in anything but Tesla (an AP2 car with FSD). But I’ve dealt enough with consumers to know things are not always quite what they seem.

When it comes to missing or hitting pedestrians, I did not walk away from the report with the belief that the consumer reports necessarily were accurate on that account. I would have to know more about the incident to form my own opinion on that. Maybe they were, maybe they weren’t, just offering a point.

Someone said it was surprising Waymo misses pedestrians. I said I feel we’d have to know more. If it misses clearly visible pedestrians crossing a street straight ahead, that would be very different from a pedestrian ”appearing out of nowhere” or changing direction unexpectedly etc...

Also, as said, the customers may not really know what the car saw and what/why it decided on such things.

There are things where the report is more fruitful and where customer reactions provide more immediately useful insight: Waymo’s hesitation, route selection issues and so forth are clearly demonstrated and those I believe. But missing pedestrians is a bit different in the sense that they can be unpredictable and a customer may have limited visibility in how the car saw or judged them.

If you have more details on the ”almost missed pedestrian” that would make me a believer, feel free to elaborate.

I guess it was me who said I was surprised the car missed pedestrians.

I don't think the sensors missed them. It has to do something with the rest of the software if it ever happened.

 

[Doggydogworld]

Surprising that it missed some pedestrians.

You can't hit 'em all Though Hollywood assures us AI will try ...

I agree with @electronblue about customer feedback. You can't take every detail literally, e.g. "I waited 6 hours on hold for Tesla service". Did a Waymo van literally sit 15 minutes before moving, or did 90 seconds feel like 15 minutes? And why didn't the rider use the phone home button?

Getting back to pedestrians, I'm not worried about Waymo going into Death Race 2000 mode. Did it barely miss illegally crossing pedestrians doing the "nudge ahead" move, or did it fail to detect them entirely? We can't tell from this report. A driver taking over and swerving to miss a cyclist sounds more serious, but again we lack details.

One thing this shows is how much Waymo is focused on the service aspect of Robotaxis vs. the technology aspect. It also shows how self-driving services are at a disadvantage vs. Uber/Lyft. Random Uber drivers can "just double park here to let me off" or whatever as long as there are no police around, but programming a robotaxi to break the law is a different matter.

 

[electronblue]

It's certainly nice to know how it happened but I still think what counts is a layperson's experience.

Sure, I wouldn’t want to dismiss it. I just put out my thoughts into our pool of communal interpretation so we have a diverse outlook on it all.

I am truly intrigued to see how Tesla’s robotaxis compare to Waymo next year.

 

[Tam]

...I am truly intrigued to see how Tesla’s robotaxis compare to Waymo next year.

Every time Tesla released a new automation feature, there have been very vocal complaints that they are so bad, so useless such as the Navigation on Autopilot, not only from in the forum but also from Consumer Reports.

I don't see any different reaction when Rotaxis will be released.

 

[electronblue]

Every time Tesla released a new automation feature, there have been very vocal complaints that they are so bad, so useless such as the Navigation on Autopilot, not only from in the forum but also from Consumer Reports.

I don't see any different reaction when Rotaxis will be released.

That will not matter if by next year Tesla robotaxis are operating without drivers in U.S. cities as Elon said they would.

 

[Daniel in SD]

Every time Tesla released a new automation feature, there have been very vocal complaints that they are so bad, so useless such as the Navigation on Autopilot, not only from in the forum but also from Consumer Reports.

I don't see any different reaction when Rotaxis will be released.

I'm trying to picture what a "useless" robotaxi would be. If you're drunk enough it doesn't really matter how well it drives. It will probably be more fun if it drives badly!

 

[Tam]

That will not matter if by next year Tesla robotaxis are operating without drivers in U.S. cities as Elon said they would.

It'll be supervised robotaxis.

We won't know how long the supervision will last until we'll get the first evaluation from our forumers

 

[humbaba]

That will not matter if by next year Tesla robotaxis are operating without drivers in U.S. cities as Elon said they would.

Elon never said that. He never even claimed there would be a Tesla ride share network of any sort next year.

What he did say was that FSD would be "feature complete" by EOY 2019 (which, if you look at the features on Tesla's site, are not sufficient for a robotaxi), "safe enough" by EOY 2020, and approved by regulators at some unknown point in the future.

The other thing he said that is often twisted/misunderstood is that there would be ~1 million cars with the 'FSD' computer by EOY 2020 with the potential to operate on a robotaxi network. That says nothing about how their owners feel about it, he was strictly talking up Tesla's increased rate of production and the fact that every car made is equipped with the new computer.

So, please, no more claims of "Elon said" and timelines for a robotaxi service. All his statements do is set a lower bound for when it would be possible (EOY 2020) and even that makes assumptions that the other requirements for a robotaxi service (beyond the listed FSD features) would be met by that time.

 

[Daniel in SD]

Elon never said that. He never even claimed there would be a Tesla ride share network of any sort next year.

What he did say was that FSD would be "feature complete" by EOY 2019 (which, if you look at the features on Tesla's site, are not sufficient for a robotaxi), "safe enough" by EOY 2020, and approved by regulators at some unknown point in the future.

I'm confused. Why would they not deploy the robotaxis in 2020 if they are "safe enough"? Waymo already has approval to deploy robotaxis.

 

[diplomat33]

 

[electronblue]

That will not matter if by next year Tesla robotaxis are operating without drivers in U.S. cities as Elon said they would.

It'll be supervised robotaxis.

We won't know how long the supervision will last until we'll get the first evaluation from our forumers

Elon never said that.

He did say it. I do grant he padded the robotaxi part with a bit more disclaimers than the Level 5 no geofence feature complete in 2019 answer which he gave.

”We expect to have the first operating robotaxis next year. With no one in them. Next year.”

”I feel very confident predicting autonomous robotaxi Teslas next year. Not all in jurisdictions because we won’t have regulatory approval everywhere but I’m confident we’ll have at least regulatory approval somewhere literally next year.”

- Elon Musk, Autonomy Investor Day

So that is the status of Tesla’s progress from those who know it best, assuming it is true. Now, if it isn’t true then all bets are of course off.

 

[humbaba]

He did say it. I do grant he padded the robotaxi part with a bit more disclaimers than the Level 5 no geofence feature complete in 2019 answer which he gave.

”We expect to have the first operating robotaxis next year. With no one in them. Next year.”

”I feel very confident predicting autonomous robotaxi Teslas next year. Not all in jurisdictions because we won’t have regulatory approval everywhere but I’m confident we’ll have at least regulatory approval somewhere literally next year.”

- Elon Musk, Autonomy Investor Day

So that is the status of Tesla’s progress from those who know it best, assuming it is true. Now, if it isn’t true then all bets are of course off.

Now your going to make me go and find a transcript. I watched investor day and I don't recall "we’ll have at least regulatory approval somewhere literally next year."

edit: here's a citation from this forum that discusses the point with sourced quotation and context: Tesla, TSLA & the Investment World: the 2019 Investors' Roundtable

 

[electronblue]

Now your going to make me go and find a transcript. I watched investor day and I don't recall "we’ll have at least regulatory approval somewhere literally next year."

edit: here's a citation from this forum that discusses the point with sourced quotation and context: Tesla, TSLA & the Investment World: the 2019 Investors' Roundtable

My quotes came straight from Elon’s mouth (via video) and the context was not that one.

It is from the presentation quoted by @diplomat33 in #756.

It is around 3:05:20 on the official video.

Elon definitely expected operational robotaxi Teslas in 2020 somewhere.

 

[Daniel in SD]

Now your going to make me go and find a transcript. I watched investor day and I don't recall "we’ll have at least regulatory approval somewhere literally next year."

edit: here's a citation from this forum that discusses the point with sourced quotation and context: Tesla, TSLA & the Investment World: the 2019 Investors' Roundtable

I feel like I'm taking crazy pills. He says that the reliability will be high enough that you can fall asleep while in the car! How is that not consistent with robotaxis?