Autonomous Car Progress

[JHCCAZ]

ps : I'm surprised Waymo is so reluctant to start robotaxi on freeways. How are they going to offer the service in LA without using highways ?

It's also interesting and a little ironic that Tesla's FSD, according to most opinion, does best on the freeway. I read comments all the time that Tesla could deploy L3 (or hands-off / eyes off or whatever you prefer) autonomy on the highway. And most competitive advanced driving assistance features are very highway-centric.

Whereas Waymo doesn't feel ready to deploy their service using freeway routing, even while operating driverless in some congested urban environments.

This probably speaks to the relative advantages and limitations of the core approaches, which as we know are quite different. I'm sure there are also differences in the perceived qualitative and quantitative risk as it affects the business model. Just one example of this would be that Waymo probably cannot afford the negative publicity of a single passenger death or serious injury - something that is more possible on a high-speed freeway, even if the statistical probability of an accident is lower than on surface streets.

 

[diplomat33]

It's also interesting and a little ironic that Tesla's FSD, according to most opinion, does best on the freeway. I read comments all the time that Tesla could deploy L3 (or hands-off / eyes off or whatever you prefer) autonomy on the highway. And most competitive advanced driving assistance features are very highway-l-centric.

Whereas Waymo doesn't feel ready to deploy their service using freeway routing, even while operating driverless in some congested urban environments.

This probably speaks to the relative advantages and limitations of the core approaches, which as we know are quite different. I'm sure there are also differences in the perceived qualitative and quantitative risk as it affects the business model. Just one example of this would be that Waymo probably cannot afford the negative publicity of a single passenger death or serious injury - something that is more possible on a high-speed freeway, even if the statistical probability of an accident is lower than on surface streets.

It is the difference between a driver assist (L2) versus driverless (L4) approach. Tesla FSD Beta has a driver in the seat that is required to supervise. So the driver can take over to prevent a highway accident. And if there is an highway accident, as we've seen with Tesla in the news, the driver is liable since FSD is L2. And we've seen in the news that Tesla just blames the driver. But with Waymo, if the car is driverless, there is no human to take over to prevent an accident. And remote assistance will likely not have time to intervene either. So the car is on its own. And if there were to be a fatal crash, the PR would be horrible as you pointed out. I mean, just imagine the headline if a passenger died in a driverless Waymo on the highway. Yikes. So Waymo needs to do their due diligence with validation before removing the safety driver on highway.

But yes, I think it shows that certain approaches can be more suited for certain ODDs. A L2 "hands-free" system can be well suited for highway ODD whereas L4 robotaxis can be well suited for city ODD.

 

[EVNow]

Just one example of this would be that Waymo probably cannot afford the negative publicity of a single passenger death or serious injury - something that is more possible on a high-speed freeway, even if the statistical probability of an accident is lower than on surface streets.

Yes - this is more to do with brand reputation risk rather than technology, IMO.

 

[Doggydogworld]

Waymo employees already have highway access now (with safety driver). So Waymo is just one step away from driverless on highways.

That's at least two steps away, and likely more based on prior experience:
- Remove safety drivers from (some) employee rides
- Offer rides with safety drivers to trusted testers
- Remove safety drivers from trusted tester rides
- Offer rides with safety drivers to the public
- Remove safety drivers from public rides

I don't think Waymo is relunctant to start robotaxis on highways.

They're extremely reluctant, by any objective measure.

It's also interesting and a little ironic that Tesla's FSD, according to most opinion, does best on the freeway.

Highways are dead simple. 100x easier than downtown streets.

Whereas Waymo doesn't feel ready to deploy their service using freeway routing, even while operating driverless in some congested urban environments.

Waymo ran 10 separate 100 mile loops, about half highway, with zero interventions. In 2009.

This probably speaks to the relative advantages and limitations of the core approaches,

Nope.

Just one example of this would be that Waymo probably cannot afford the negative publicity of a single passenger death or serious injury - something that is more possible on a high-speed freeway, even if the statistical probability of an accident is lower than on surface streets.

You nailed it. Waymo is extremely risk-averse because one death can sink them (as it did Uber's self-driving project). Tesla simply blames the decapitee.

 

[Jeff N]

For what it’s worth, every time I’m on US-101 in the SF Bay Area I end up seeing a Waymo in the slower lanes. I don’t tend to drive much further south of SF than San Mateo so I don’t know if they are testing drives to SFO or all the way down to the South Bay.

 

[diplomat33]

Highways are dead simple. 100x easier than downtown streets.

Simpler yes but more risky. As discussed, the reason Waymo is not doing driverless on highways yet is because of risk, not technology.

Waymo ran 10 separate 100 mile loops, about half highway, with zero interventions. In 2009.

Yes, Waymo had the tech to do highways in 2009. But there is a big difference between doing 500 miles with zero intervention and having the reliability that you could trust the car to be driverless 24/7 and not cause an accident. That is why it is not a tech issue, it's a validation and risk issue.

 

[EVNow]

As discussed, the reason Waymo is not doing driverless on highways yet is because of risk, not technology.

The important thing to understand here is that - its not the risk of accident as such, but reputational risk because that one accident that can be fatal (to the passenger / other driver ... and to reputation).

 

[diplomat33]

The important thing to understand here is that - its not the risk of accident as such, but reputational risk because that one accident that can be fatal (to the passenger / other driver ... and to reputation).

Correct but accident risk is related to reputational risk. The severity of an accident on highways is potentially higher due to the higher speeds. So if an accident happens on the highway, the reputational risk is higher.

 

[EVNow]

Correct but accident risk is related to reputational risk. The severity of an accident on highways is potentially higher due to the higher speeds. So if an accident happens on the highway, the reputational risk is higher.

What I mean is … per mile stats for city is higher than highway. So, the accident rates (including fatality) are worse for city than highways.

 

[diplomat33]

What I mean is … per mile stats for city is higher than highway. So, the accident rates (including fatality) are worse for city than highways.

Correct but I was talking about severity of accident. City driving has more accidents per mile but they tend to be less severe. Highway has fewer accidents per mile but they tend to be more severe.

 

[powertoold]

Waymo ran 10 separate 100 mile loops, about half highway, with zero interventions. In 2009.

Sure, but did those miles include lane changes, merging, exiting, and heavy traffic situations?

Every time I saw those Waymos driving on highways years ago, it was in light traffic, and I never saw them doing anything slightly complex.

I'm likely wrong though

 

[DanCar]

In 2009, Waymo always had safety drivers, so it was L2. My internal source told me that Waymo hasn't done freeways, not for tech reason(s), but for legal reasons. Lane merging is tricky and almost always the fault of the person trying to merge in. On crowded freeway a waymo can attempt a merge, someone can seemingly allow the car to move in, then hit the waymo , and from a strictly legal perspective it would be Waymo's fault. I'm not overly confident my internal info is accurate, but that is what I heard, perhaps 3 years ago.

 

[powertoold]

Lane merging is tricky and almost always the fault of the person trying to merge in. On crowded freeway a waymo can attempt a merge, someone can seemingly allow the car to move in, then hit the waymo , and from a strictly legal perspective it would be Waymo's fault. I'm not overly confident my internal info is accurate, but that is what I heard, perhaps 3 years ago.

That sounds odd, especially for an AV with cameras all around.

I haven't found any video or evidence of how well or poorly the Google self-driving project cars drove highway. If any of you have any promotional or raw video of Google or Waymo driving highway situations autonomously, please share.

 

[JB47394]

If any of you have any promotional or raw video of Google or Waymo driving highway situations autonomously, please share.

I searched YouTube for "highway Waymo" and got several hits for big rig trucks. I also saw a 5 year old video showing a Waymo van on the highway. So they've apparently been testing for a while.

 

[powertoold]

I searched YouTube for "highway Waymo" and got several hits for big rig trucks. I also saw a 5 year old video showing a Waymo van on the highway. So they've apparently been testing for a while.

Sure, but do we see how well it works or not, like inside the car and visualizations of it dealing with highway situations? The videos are very sparse compared to city streets.

I also consider trucking to be a different category of highway driving, so hard to compare with normal minivans or SUVs.

 

[Bitdepth]

Sure, but do we see how well it works or not, like inside the car and visualizations of it dealing with highway situations? The videos are very sparse compared to city streets.

I also consider trucking to be a different category of highway driving, so hard to compare with normal minivans or SUVs.

In 2019 they used to do highways as part of the consumer testing in AZ but it got taken out because it was too cautious to merge or exit the highway so they removed it. There used to be some videos but they are old now so i can't find them. Now they are doing trucking exclusively on highway, but the robotaxi is testing highways for consumer rides with employees only.

 

[DanCar]

Waymo failing to merge video, from 2018.

https://twitter.com/x/status/990683818825736192

 

[Doggydogworld]

Sure, but did those miles include lane changes, merging, exiting, and heavy traffic situations?

The goal in 2009 was to show a path to safe autonomous driving. Without safety everything else is just BS. "Features" are easy to add, but mostly serve to excite fanboys (and generate GAAP revenue for Tesla, lol).

Waymo posted some clips from the 2009 tests. Most are sped-up and jerky. The second one is 1x speed, but super boring. Generally speaking, they:
- obey speed limits, so cars stream by on both left and right
- only change lanes when needed to exit, or avoid exiting
- pretty much yield to everyone

The first clip shows a bit of traffic after 1:15. Also some extreme yielding in a construction zone around 2:20. It's been years since I watched the clips, so I know there are a few other interesting scenes but don't recall details.

 

[EVNow]

Waymo failing to merge video, from 2018.

https://twitter.com/x/status/990683818825736192

Waiting for @Bladerskb to tell us why this shows Tesla FSD is bad

 

[JB47394]

Waymo has all those LIDAR hanging off it, screaming that this is an unusual car and should be treated as such. It would be nice if Teslas came with something that other drivers intuitively understood to mean that the car is using automation software. I know that some people put bumper stickers on their Teslas, but stickers are only visible at short distances, from behind, and what they say isn't true when the driver is actively controlling the car.

I can imagine a NHTSA mandate that cars running automation software must clearly indicate it to other drivers. How about a taxi light on top that can light up with L1 to L5? Or some roof rack-mounted light bar with 5 bands that light up depending on the level of driving automation?

I just think that being able to communicate to other drivers that "It's not me, it's my car" is something that should happen.