Autonomous Car Progress

[ItsNotAboutTheMoney]

We have them in Rhode Island, though I'm not sure anyone knows how they are supposed to work.

There's multiple-lane rotaries in Augusta, ME. Rotary law is simple here. Yield at entrance, allow traffic to change lanes to exit. Still, I know people who take alternative routes, particularly to avoid the larger of the two rotaries in the city.

The smaller rotary was made a roundabout several years ago, with a spiral design that means if you get into the correct lane as you enter, you don't have to cross traffic to exit.

Exit 113 of I-95 north of Augusta was changed a few years ago when a new hospital campus was built. Previously the exit was eastbound only, but it was changed to allow westbound traffic, with a rotary added each side of the overpass. Negative for eastbound traffic exiting I-95S, who now have to cross the two rotaries before getting back up to speed.

I don't think rotaries are particularly complicated for autonomy, but they are a good test.

 

[banned-66611]

Have not any of the beta testers started on a coast-to-coast test yet?

Obviously impossible because it doesn't have self-charging or self-cleaning yet.

 

[Soda Popinski]

Obviously impossible because it doesn't have self-charging or self-cleaning yet.

Judging that it's still Level 2 at this point, requiring driver intervention every once in a while, I don't see why self-charging would be a pre-requisite, as there will still be a driver in the car.

 

[drtimhill]

The smaller rotary was made a roundabout several years ago, with a spiral design that means if you get into the correct lane as you enter, you don't have to cross traffic to exit.

 

[daktari]

Obviously impossible because it doesn't have self-charging or self-cleaning yet.

I would give it a "great" rating even if the driver need to back up and plug in manually. Without interventions between SuCs would be great.

 

[AlanSubie4Life]

Have not any of the beta testers started on a coast-to-coast test yet?

Have you watched any of the videos?

 

[Daniel in SD]

Have not any of the beta testers started on a coast-to-coast test yet?

What would be the point? I'm sure someone will do it when there's a good chance of making it without disengagements. Really it should be done without Interstate Highways to make it interesting.
Doing it without a driver is probably not going to happen any time soon so worrying about automated charging/cleaning seems a little premature.

 

[AlanSubie4Life]

worrying about automated charging/cleaning seems a little premature.

I'm sure early next year Tesla will announce they will soon have Supercharger attendants, to monitor arrival of the summoned vehicles, and plug them in and clean any sensors. No need for automation. Jobs, jobs, jobs. Tesla Pit Crew.

 

[Daniel in SD]

I'm sure early next year Tesla will announce they will soon have Supercharger attendants, to monitor arrival of the summoned vehicles, and plug them in and clean any sensors. No need for automation. Jobs, jobs, jobs. Tesla Pit Crew.

Probably, as Elon said, humans are underrated. The complexity of automated supercharging may not be worth the cost. Also, the robotaxis are going to need to be cleaned on the inside too! A robot that cleans as well as a human might be almost as valuable an invention as a robot car and probably even more difficult to achieve.

 

[daktari]

Have you watched any of the videos?

Yes, and sadly enough not impressed.
Alert me when someone does coast-to-coast, then its interesting.

 

[diplomat33]

I think a coast to coast FSD drive has a certain appeal because it would illustrate how Tesla's FSD works everywhere in the US, especially considering Tesla's "generalized FSD" approach. There is a cool factor that you could get into your Tesla in LA and it could drive you all the way to NY (hopefully) without interventions. But a coast to coast FSD drive would actually not be that meaningful. It would be mostly highway driving which is easy to do, especially if you drive at times when traffic is light. And it is only 3,000 miles which is a tiny fraction of how many miles FSD needs to be able to do reliably. Most autonomous driving companies had the ability to do a coast to coast FSD drive years ago.

 

[drtimhill]

Have not any of the beta testers started on a coast-to-coast test yet?

Not really much point, as that would mostly be freeway driving using NoA, which hasn't changed in the FSD beta (it's still using the old AI stack).

 

[Daniel in SD]

Once FSD is ready for final verification Tesla should drive all 2.7 million miles of paved road in the United States. Obviously there would be disengagements and accidents when driving that distance but it would probably provide enough data to prove safety.

 

[banned-66611]

I question the safety of a coast to coast drive. Can a human reliably monitor it for 50+ hours? I think it would be nearly impossible to stay attentive for that long.

They should be using professional test drivers in 1 hour stints, with eye tracking cameras.

 

[diplomat33]

Here is a very cool in-depth video of how Zoox does teleguidance of their autonomous cars. The video is very informative:

Note that the teleoperators do not directly drive the car. They merely provide suggestions to the car but the car never disengages from autonomous driving. They can draw a recommended path if the autonomous car is not sure what path to take. For example, if the autonomous car is not sure how to navigate a construction zone, the teleoperator can draw a path for the car to take. And they can label objects if the car is not sure how to respond to an object. For example, we see that the car is confused by a stopped delivery truck. The teleoperator labels the truck as a blocking obstacle and the car then automatically tries to go around it. The AI learns from these instances to require less guidance in the future. Right now, Zoox autonomous cars require guidance from a teleoperator about 1% of the time.

Also note that the perception visualization is incredibly detailed and accurate.

 

[ThomasD]

How do they prevent signal interference or signal loss? Do they have a way to keep the cameras clean at all times?

 

[diplomat33]

How do they prevent signal interference or signal loss?

It does not matter. The cars don't require teleguidance. They still drive autonomously even with teleguidance. And teleguidance is only 1% of the time.

But I assume the teleguidance is communicated over 4G or 5G. And they drive around the city so they probably have a strong signal most of the time.

Do they have a way to keep the cameras clean at all times?

I know Waymo has self-cleaning cameras. Zoox might have something similar.

 

[banned-66611]

1% of the time is quite a lot. Say each event lasts for 20 seconds, that's one intervention every half hour.

 

[diplomat33]

1% of the time is quite a lot. Say each event lasts for 20 seconds, that's one intervention every half hour.

Well, keep in mind that Zoox's autonomous driving is not as advanced as say Waymo's. For comparison, Zoox's disengagement rate is about 7x worse than Waymo's.

 

[Daniel in SD]

1% of the time is quite a lot. Say each event lasts for 20 seconds, that's one intervention every half hour.

The CTO of Cruise said every 5-10 miles. That would be 2-4 times per hour (average speed 20mph) or 1-2% of the time if the events lasts 20 seconds.