JALOPNIK: Uber Is Here To Remind You That Autonomous Vehicles Aren't Even Close

[Hoppi]

https://jalopnik.com/uber-is-here-to-remind-you-that-autonomous-vehicles-are-1845207289

Uber Is Here To Remind You That Autonomous Vehicles Aren't Even Close

Uber, like its competitors Waymo, Zoox, Cruise and others, has been trying and failing to get autonomous cars up and running for years now. The business opportunity — no more of that having-to-employ-drivers nonsense — is obvious, but it’s funny how everyone in the self-driving business seems to have stopped talking about self-driving.

They haven’t given up, of course. But remember in 2018 when Uber said that it wanted to get you in a driverless car by the end of that year? Or when Cruise said it would test self-driving cars in New York City? Or what about when Volvo said it would have a fully autonomous car by next year?

Those timelines haven’t quite worked out, unless Volvo really is ready to go next year with a Level 5 autonomous car. I’m not holding my breath! The reasons for this delay aren’t a mystery: Creating a car able to safely deal with every real-world contingency may, in fact, be close to impossible, but everyone is still trying for now.

That includes Uber, which, like Cruise before it, has been the subject of a deep dive from The Information. We learn, for example, that the fatal 2018 crash in Arizona seems to have shaken the company, even as no one aside from the safety driver suffered any apparent consequences from it. We also learn that Uber isn’t at all close to getting over the line with an autonomous car.

Take this snippet:

[In 2019], Uber’s vehicles had also been having what the company refers to as a bad experience — such as a sudden jerk or a potentially dangerous movement — every one-third of a mile on average. The company’s leaders had hoped the figure would fall to just one bad experience every 10 miles by fall last year, said a person with direct knowledge of the goal. One person who worked on the effort told The Information they felt the prototypes were better than the data indicated.
Uber couldn’t even get the prototype to drive a one-mile stretch between the unit’s two Pittsburgh offices, with the goal of shuttling employees back and forth. Software leaders said they gave up trying because automating the route for internal use wouldn’t help Uber develop software that it could apply to a broader array of routes. But other managers said the failure sent a message that Uber just couldn’t pull it off.

Or this one (emphasis mine):

And in July this year, [Uber self-driving chief Eric Meyhofer] and his team gave a presentation in which they outlined their road map. They said they had selected 10 or so miles of public roads in Pittsburgh and Washington, D.C., in which to test vehicles, and had plans to do the same for San Francisco, Toronto and Dallas in the future. They spent time discussing the new rNA software, which by the end of September was supposed to power a prototype—a Volvo SUV with Uber’s tech installed—in public road tests. One current employee and one former employee said that is likely to slip to October.
At the time of the July presentation, the company’s data showed that Uber was only 3.9% of the way toward figuring out how to test the vehicle’s readiness to handle potential situations arising on the road.
In a four-way stop, for instance, a vehicle must be able to detect pedestrians, obey the stop sign, let a pedestrian pass and so on, with each one of those items representing a situation, or test case. The slide projected that a prototype would need to pass 28,778 test cases for those road scenarios by the end of next year—when Uber wants a driverless vehicle to be on the road in Pittsburgh, with no human backup.

I recommend reading the whole (paywalled) story here. I would quibble only a bit with the headline, which says that Uber has “wasted” $2.5 billion on its self-driving efforts. The fact is, if Uber has wasted billions in research and development money on self-driving, so has everyone else.

That’s the nature of spending money on research and development. Sometimes it works out, sometimes it doesn’t. So far, none of it’s worked out, if you take working out to be producing an autonomous car capable of dealing with every contingency. It’s gonna take many more billions to get there — if we can ever achieve that.

 

[Hoppi]

Not sure what the other companies developing autonomous driving are thinking but this article makes it look like Uber is nowhere near getting it to work.

Compare that to Elon's pronouncements of being close to being able to have FSD out.

If Tesla is the only company with FSD for any significant amount of time, that will be a serious advantage over other vehicles / ride-shares.

 

[ajdelange]

That would be the case but Tesla isn't going to have FSD in the foreseeable future either if it ever does. It is an impossible problem.

 

[Blue Steel]

That would be the case but Tesla isn't going to have FSD in the foreseeable future either if it ever does. It is an impossible problem.

I tend to agree. I'm really curious how much of a step forward the big rewrite will be. FSD needs a massive step forward if it's ever going to be realized.

I was listening to the Ride The Lightning podcast today and Ryan made an interesting point. A few years ago Elon talked about a $25k Tesla but then said that it wasn't needed because full autonomy is on the way. However on battery day the $25k Tesla is back on the table.

 

[Hoppi]

I tend to agree. I'm really curious how much of a step forward the big rewrite will be. FSD needs a massive step forward if it's ever going to be realized.

I was listening to the Ride The Lightning podcast today and Ryan made an interesting point. A few years ago Elon talked about a $25k Tesla but then said that it wasn't needed because full autonomy is on the way. However, on battery day the $25k Tesla is back on the table.

Hadn't heard about Elon saying that. I was thinking though while responding on some other threads about how people with tighter incomes would first get exposed to electric vehicles. Rideshare/taxi service might be first as long as FSD is around to make the ride cheap enough to compete with buying your own vehicle. But with ICE vehicle prices dropping as we switch to BEV it might be very difficult to make that work. At some point, the maintenance and gas alone will make ICE vehicles cost-prohibitive. But till then, either there needs to be a selection of very inexpensive BEV's or FSD robo-taxis need to be all over the place and cheap enough to ride regularly.

This also makes me wonder when the price of rideshare/taxi will start to compete with public transport. At some point, the convenience and cost of rideshare will be improved enough that public transport will probably collapse.

It's funny, but about the time that Falcon Heavy first flew, Elon was saying that he intended to have 1 million people on mars by 2035. At the same time, I was working in a building in Bellevue WA overlooking the building of a light rail station that would be part of a network into Seattle. The furthermost extension of which was planned for Lynwood (20 miles away) by 2036. I was laughing about Elon having a million people on Mars before I could ride light rail to Lynwood.

I do wonder if large public transportation projects are pretty much outdated now. Long distance projects are probably still a thing (hyperloop between large cities anyone?) though.

 

[Crissa]

Our political system is designed to have lots of chokeholds, which is why our big projects are so expensive.

But that one-per-1/3-mile stat is worse than they were the year they killed someone. Uber created the situation the driver was in, an impossible task. ...While they needed millions of miles of data like Tesla is harvesting and Way o is generating.

Uber is really an example of how not to do... pretty much everything they do.

-Crissa

 

[Jhodgesatmb]

That would be the case but Tesla isn't going to have FSD in the foreseeable future either if it ever does. It is an impossible problem.

You are right of course, but the problem is much more impossible for human drivers. Tesla has already riven to be an order of magnitude safer than people in terms of accident rates, so the question is never going to be whether it is perfect but whether it is good enough. That is a difficult question to answer in a vacuum but if we always put it up next to human drivers we might get a better idea of when it is good enough. That is what authorities need to do. I drive into situations every single day, and more all the time, where people are driving against the law, not paying attention, etc., and there is virtually no way to predict how to go through an intersection safely. We just take a calculated risk, and our powers of calculation aren’t that good. I guess I would prefer a car trying it’s hand at an impossible situation than a person, but in driving with autopilot it still has some way to go on lots of streets and conditions to convince me to use it off the highway.

 

[TheLastStarfighter]

Our political system is designed to have lots of chokeholds, which is why our big projects are so expensive.

But that one-per-1/3-mile stat is worse than they were the year they killed someone. Uber created the situation the driver was in, an impossible task. ...While they needed millions of miles of data like Tesla is harvesting and Way o is generating.

Uber is really an example of how not to do... pretty much everything they do.

-Crissa

Uber Eats is amaaazing.

 

[ajdelange]

question is never going to be whether it is perfect but whether it is good enough. That is a difficult question to answer in a vacuum

It's a difficult question to answer - period. Musk got into that a bit in one public appearance (not the shareholders' meetin) in which he asked how many 9s were enough and commented that they are at the point where the extra 9s are really hard to get.

 

[Jhodgesatmb]

It's a difficult question to answer - period. Musk got into that a bit in one public appearance (not the shareholders' meetin) in which he asked how many 9s were enough and commented that they are at the point where the extra 9s are really hard to get.

You of all people know what I meant, that deciding what is good enough requires a baseline and a metric and threshold. I suspect that the baseline will be something like the number of accidents per 1,000,000 driven miles. I do not now whether they (who is they?) will break it out by driving type (city, highway, suburbs, rural, mountains, etc.), driving conditions (clear, rain, snow, fog), road conditions (dry, wet, flooded, muddy, slushy, icy, frozen, etc.), or just totaled over all of the above. One thing for certain is that what Elon Musk considers FSD, what the NHTSA thinks is Level 5 Autonomous Driving, and what regulators decide is public policy for autonomous driving are different, let alone what ‘we’ think. I believe whatever it is now it will be much better when CTs roll off the assembly line.

 

[MEDICALJMP]

That would be the case but Tesla isn't going to have FSD in the foreseeable future either if it ever does. It is an impossible problem.

Guessing you did not opt for FSD in your reservation.

 

[ajdelange]

You of all people know what I meant,

Of course. Just emphasizing how difficult the question is.

 

[ajdelange]

Guessing you did not opt for FSD in your reservation.

Yes, I did.

 

[Crissa]

Guessing you did not opt for FSD in your reservation.

Yes, I did.

Is English not your first language? This is one case where two positives cancel each other out. 'Yes, you did opt out of FSD' or 'Yes, you did opt for FSD'?

-Crissa