Autonomous Car Progress

[mspohr]

Can these Trucks adapt to changing Weather and Road conditions and adjust speed accordingly? If the Trucks need assistance do they use remote drivers or do they send people out to rescue the Trucks? If they use remote help does that person need a CDL? Is Kodiak testing in Snow?

I drove my Model S in a snow storm today. Got a message on the screen that one or more cameras were blocked.
Would FSD work?

 

[voyager]

VTPI published a recent overview where all 'driverless' considerations, speculations, scenarios and challenges are
neatly ordered: https://www.vtpi.org/avip.pdf

"some aspects"

Vehicles last longer, cost more, impose larger external costs, and are more highly regulated than most other consumer goods.
As a result, vehicle technologies take longer to penetrate markets than most other sectors. It will probably take decades for
autonomous vehicles to dominate new vehicle purchases and fleets, and some motorists may resist using them.

An important planning issue is whether autonomous vehicles will increase or reduce total vehicle travel and associated traffic problems.

Road space is a scarce and valuable resource. Horizontal equity requires giving priority to space-efficient vehicles

Private autonomous vehicles are likely to have low occupancy rates. As previously described, without efficient road pricing
it will often be cheaper for motorists to program their autonomous cars to circle the block or return home, to avoid paying
for off-street parking, which will contribute even more to traffic congestion.

Reports are primarily oriented toward investors, and so focus on autonomous vehicle’s sales potential, but most policy
and planning decisions depend on the portion of autonomous vehicles in the fleet and in total vehicle travel.

This analysis suggests that it will be at least 2045 before most vehicles are autonomous, and longer before they are affordable.

Autonomous vehicles are likely to support some equity goals but contradict others.

 

[KArnold]

I'm more familiar with Aurora, a Pittsburgh-based company working on autonomous 18-wheelers. They are using safety drivers today but plan to do driverless between Dallas and Houston this year. And they will.

Can these Trucks adapt to changing Weather and Road conditions and adjust speed

Yes they are very weather aware and adjust accordingly. I don't know what snow testing they have done given the route. I'd expect they are not yet ready for snow or ice at this point in evolution, just like cars. They do handle rain quite well.

If the Trucks need assistance do they use remote drivers or do they send people out to rescue the Trucks?

Remote support. Don't know about CDL - good question.

 

[diplomat33]

CA DMV permits now include area and ODD. As of Jan 11, 2024, 3 companies have driverless permits:

Mercedes-Benz

Area:
· California freeways and highways in:
– Bay Area
– Los Angeles
– Sacramento
– San Diego
· Interstate 5
· Interstate 15 connecting Los Angeles area to Nevada

ODD:
Daytime. Sufficient weather conditions
Excludes: Flooded highways, heavy smoke, heavy dust, and heavy fog, snowstorms
· Speed 40 mph

Note: This technology as a traffic jam pilot corresponds to SAE Level 3, which means that not only must there always be a driver in the driver’s seat and be able to take over at any time, the vehicle automatically only keeps the lane in a traffic jam, may only do this on a highway, and always requires a so-called lead vehicle, i.e. a vehicle in front, which is controlled by a human and which the Mercedes-Benz then follows.

Nuro

Area:
· San Mateo County
· Santa Clara County

ODD: All times of day and night
· Fair weather conditions:
– dry or wet pavement/asphalt
– light rain
– light to moderate fog
· Speed 25-35 mph


Waymo

Area:

· San Francisco County
· San Mateo County
· Santa Clara County
· Portions or all of the
following cities:
– Atherton
– Belmont
– Brisbane
– Burlingame
– Colma
– East Palo Alto
– Foster City
– Hillsborough
– Los Alto
– Los Alto Hills
– Menlo Park
– Millbrae
– Mountain View
– Palo Alto
– Portola Valley
– Redwood City
– San Bruno
– San Carlos
– San Mateo
– South San Francisco
– Sunnyvale
– Woodside
· Los Angeles County
· Portions or all of the
following cities:
– Bell
– Bell Gardens
– Beverly Hills
– Carson
– Commerce
– Compton
– Cudahy
– Culver City
– El Segundo
– Gardena
– Hawthorne
– Huntington Park
– Inglewood
– Lawndale
– Long Beach
– Los Angeles
– Lynwood
– Manhattan Beach
– Maywood
– Paramount
– Redondo Beach
– Santa Monica
– South Gate
– Torrance
– Vernon
– West Hollywood

ODD: All times of day and night
· Inclement weather, rain, and fog
· Speed 65 mph

California Licenses for Autonomous Vehicles in Detail

On January 11, 2024, the California Department of Motor Vehicles (DMV) made the details of the licenses issued for driverless autonomous driving more easily accessible and clearer. The DMV website …

thelastdriverlicenseholder.com

 

[KArnold]

Or Ohio, Washington, Maine (and other areas, I'm sure), the entire state. There are plenty of areas capable of testing L4/L5.

https://www.motortrend.com/features...-self-driving-driverless-cars-vehicles-legal/

 

[willow_hiller]

ODD: All times of day and night
· Inclement weather, rain, and fog
· Speed 65 mph

It would be interesting to see whether there are some additional technical details beyond this description. Because just the above makes it sound like Waymo's ODD is not restricted by weather at all, but there are still some increment weather conditions they cannot drive in.

For e.g. dense fog last April caused some Waymo drivers to enter their minimal risk condition: https://www.sfchronicle.com/bayarea/article/san-francisco-waymo-stopped-in-street-17890821.php

 

[diplomat33]

It would be interesting to see whether there are some additional technical details beyond this description. Because just the above makes it sound like Waymo's ODD is not restricted by weather at all, but there are still some increment weather conditions they cannot drive in.

Yeah. Waymo's CA DMV permit certifies them to drive in rain and fog in general but it will depend on how dense the rain and fog is. The rain or fog could be bad enough that Waymo does not drive in them.

Part of it might that Waymo is continually improving their system to handle rain and fog better. So what used to be fog so dense the Waymo Driver had to stop may now be just fine for the Waymo Driver to drive in. I know Waymo has made a lot of progress now and it is able to drive in most dense rain and fog now.

 

[willow_hiller]

Yeah. Waymo's CA DMV permit certifies them to drive in rain and fog in general but it will depend on how dense the rain and fog is. The rain or fog could be bad enough that Waymo does not drive in them.

As AV's become more advanced, I think having clearly defined ODD will become more difficult. The above shows we're pretty much already at the point where the answer to the question "What weather conditions can Waymo driver not handle?" is "Waymo driver can handle all weather conditions except for the ones it decides it cannot handle."

I've had arguments in other threads where people say that Tesla's FSD Beta can never achieve a level of autonomy higher than L2 without a strictly defined ODD; but I maintain that as long as a system is advanced enough to determine when it cannot safely drive by itself, and perform a timely handover routine or minimal risk condition for L3 and L4 respectively, the ODD becomes defined purely by the probabilistic output of the system. You can draw a rough circle around the types of situations that the system cannot handle, but you lose the ability to put it in a neat box.

 

[diplomat33]

As AV's become more advanced, I think having clearly defined ODD will become more difficult. The above shows we're pretty much already at the point where the answer to the question "What weather conditions can Waymo driver not handle?" is "Waymo driver can handle all weather conditions except for the ones it decides it cannot handle."

ODD always needs to be clearly defined. So you cannot just say "Waymo driver can handle all weather conditions except for the ones it decides it cannot handle." But I don't think you are going to define the ODD as detailed as you seem to think. For example, you don't need to specify the exact cubic inches of rain fall that the AV can handle. It is enough to say the ODD is "light rain" or "moderate rain". But you do need to say that. Of course, the AV will have software that will measure how degraded the perception is and what the threshold is to where it decides to stop or pull over. That threshold will be well defined in the software.

I've had arguments in other threads where people say that Tesla's FSD Beta can never achieve a level of autonomy higher than L2 without a strictly defined ODD; but I maintain that as long as a system is advanced enough to determine when it cannot safely drive by itself, and perform a timely handover routine or minimal risk condition for L3 and L4 respectively, the ODD becomes defined purely by the probabilistic output of the system. You can draw a rough circle around the types of situations that the system cannot handle, but you lose the ability to put it in a neat box.

By definition, L4 is autonomy in a limited ODD. So you must define the limits of the ODD to do L4. So by definition, FSD beta would need to define the ODD if it wanted to be L4. it cannot say "I am L4 and I will just pull over if I can't handle something". But the ODD is what the system is designed to operate in, not every single condition that the system can handle. So if you design your L4 to drive on highways, then highways are part of the ODD, even if the L4 might need to pull over under certain highway conditions. If you design your L4 to drive in light rain, then light rain is part of the ODD, heavy rain would be outside the ODD if your L4 is not designed to operate in heavy rain etc...

Put simply: L4 will pull over if it can't handle something inside its ODD or if it is going to leave its ODD, but you still need to define the ODD since that defines where and when you can use the L4. For example, can I engage L4 on rural roads or highways? Can I engage L4 in light rain or no rain at all? Can I engage L4 at night? Can I engage L4 at speeds above 40 mph? Those are the type of questions the ODD needs to answer.

The ODD is basically when the little grey wheel shows up that FSD beta is available. When you see that little grey wheel, you are inside the ODD. For L4, Tesla would need to define the ODD (road type, weather, time of day, etc) when they want the grey wheel to pop up that you can turn on L4 or not. That's because you would not the user to turn on a L4 system where they don't need to supervise on a road or weather condition or time of day etc that the L4 is not designed to operate in.

 

[willow_hiller]

By definition, L4 is autonomy in a limited ODD. So you must define the limits of the ODD to do L4. So by definition, FSD beta would need to define the ODD if it wanted to be L4. it cannot say "I am L4 and I will just pull over if I can't handle something". But the ODD is what the system is designed to operate in, not every single condition that the system can handle. So if you design your L4 to drive on highways, then highways are part of the ODD, even if the L4 might need to pull over under certain highway conditions. If you design your L4 to drive in light rain, then light rain is part of the ODD, heavy rain would be outside the ODD if your L4 is not designed to operate in heavy rain etc...

Is there any difference between a design intent, and something that has been intentionally designed?

Because I don't know what else to call Waymo entering a minimal risk condition upon detecting fog. Either the programmers of Waymo intentionally designed the system to cease the driving task when fog is detected, or it happened unintentionally. I don't think it happened unintentionally, so the system was designed to not operate in fog. By extension, fog should not be included as a part of Waymo's ODD, but they have explicitly included it.

This use of "design intent" is more coming across as "design aspiration" instead. Aspirationally, Waymo wants their system to be able to drive in all inclement weather. But by virtue of how they have designed the system as it actually exists, they have intended the system to cease the driving task and enter the minimal risk condition upon detecting it.

 

[diplomat33]

Is there any difference between a design intent, and something that has been intentionally designed?

Design intent is what you plan for the system to do. Intentionally designed would refer more specifically to the effort you put into making the system do something.

Ex: design intent is L4 because you plan for the system to be L4 and you have a design on paper to do L4. Intentionally designed would refer to all the work you do to make the system reach L4.

Because I don't know what else to call Waymo entering a minimal risk condition upon detecting fog. Either the programmers of Waymo intentionally designed the system to cease the driving task when fog is detected, or it happened unintentionally. I don't think it happened unintentionally, so the system was designed to not operate in fog. By extension, fog should not be included as a part of Waymo's ODD, but they have explicitly included it.

I think you are missing the reliability part. A system can be designed to work in a given ODD. That does not automatically mean that it works 100% of the time. Waymo includes fog in the ODD because it is designed to work in fog and has been tested to work in fog with sufficient reliability. So maybe Waymo handles fog safely 99% of the time. The other 1% of the time, it enters a MRC. What you saw in the video when it entered a MRC was that instance when the Waymo decided it could not safely proceed but there are plenty of other instances where Waymo handles fog just fine. Waymo does not enter MRC every single time it enters fog, only in certain cases.

Put simply, if you design your system to handle a given ODD and it works reliably enough, you might decide that is good enough to include in the ODD. It is up to the manufacturer to decide if the AV is reliable enough and safe enough in a given ODD to include it or not. Waymo includes fog in the ODD because they deem it is safe enough in fog.

This use of "design intent" is more coming across as "design aspiration" instead. Aspirationally, Waymo wants their system to be able to drive in all inclement weather. But by virtue of how they have designed the system as it actually exists, they have intended the system to cease the driving task and enter the minimal risk condition upon detecting it.

No, it is not about aspiration at all. Waymo designed the system to work in a given ODD and they tested that it works in that ODD with acceptable reliability and safety. That is why they include that ODD because the system has been tested to work reliably enough.

Remember there was a time when fog was not in the ODD because Waymo could not handle fog reliably. Waymo added fog to the ODD after they improved the system and tested it in fog, and deemed that it now does handle fog reliably enough.

Again, Waymo does not enter MRC every time it detects fog. It is not designed to automatically reach MRC when it detects fog. It only reaches MRC if the particular fog conditions warrant a MRC. So most of the time, Waymo will handle fog just fine. It only enters MRC if a particular fog condition happens to warrant a MRC.

To reiterate:
- You design the system to work in a given ODD
- You test the system that it works reliably in that ODD.
- If the reliability is good enough, you include it in the ODD.
- You design the system to reach MRC if the conditions warrant it.
- The system can work reliably in a given ODD and still reach MRC in rare occasions because of a specific case.

For example: the AV is designed to work in fog. It works reliably 99% of the time in fog. So 99% of the time, it does not need to reach MRC in fog and can drive in fog just fine. But in one instance, maybe it happens to encounter super dense fog then it reaches MRC in that given case. It reaches MRC because of that specific case where the fog was too dense not because it can't handle fog at all.

 

[willow_hiller]

I think you are missing the reliability part. A system can be designed to work in a given ODD. That does not automatically mean that it works 100% of the time. Waymo includes fog in the ODD because it is designed to work in fog and has been tested to work in fog with sufficient reliability. So maybe Waymo handles fog safely 99% of the time. The other 1% of the time, it enters a MRC. What you saw in the video when it entered a MRC was that instance when the Waymo decided it could not safely proceed but there are plenty of other instances where Waymo handles fog just fine. Waymo does not enter MRC every single time it enters fog, only in certain cases.

Put simply, if you design your system to handle a given ODD and it works reliably enough, you might decide that is good enough to include in the ODD. It is up to the manufacturer to decide if the AV is reliable enough and safe enough in a given ODD to include it or not. Waymo includes fog in the ODD because they deem it is safe enough in fog.

I think you're conflating reliability of the system with the intensity of the operational environment. Notice in the photo of the incident in April, every single Waymo came to a stop in that fog:

The reliability for fog of that intensity was 0%. It does not seem like there is a chance that Waymo can operate in that fog.

Notice in the other ODDs, other AV companies qualified the intensity of the environment. Mercedes said it excluded "heavy fog," Nuro included "light to moderate fog." Waymo could have specified that their system can handle light to medium fog, but not dense fog. But instead they decided to include all inclement weather in their ODD.

 

[diplomat33]

I think you're conflating reliability of the system with the intensity of the operational environment. Notice in the photo of the incident in April, every single Waymo came to a stop in that fog:

View attachment 1008362

The reliability for fog of that intensity was 0%. It does not seem like there is a chance that Waymo can operate in that fog.

Notice in the other ODDs, other AV companies qualified the intensity of the environment. Mercedes said it excluded "heavy fog," Nuro included "light to moderate fog." Waymo could have specified that their system can handle light to medium fog, but not dense fog. But instead they decided to include all inclement weather in their ODD.

You are assuming all the Waymos stopped because Waymo cannot handle fog at all. That is false. Waymo could not handle that particular fog. There is a difference.

And that video might have been from before Waymo added fog to ODD since I think you said it was from last April. That was awhile ago. I would need to check. I know fog is in the ODD now and Waymo can handle fog now.

 

[willow_hiller]

You are assuming all the Waymos stopped because Waymo cannot handle fog at all. That is false. Waymo could not handle that particular fog. There is a difference.

I didn't say all fog. I said fog of that intensity. I don't see any difference between fog of that intensity in April 2023, and fog of the same intensity if it were to occur again.

The facts of the matter are that in April 2023, the fog was dense enough that every Waymo driving in it stopped. So we know that at that point in time, fog of that intensity was not within Waymo's ODD.

They put out a blog in August where they say "our Driver is now equipped to handle even heavier levels of fog than before." But it says "heavier fog" not "heavy fog."

I still think it's a problem that Waymo is claiming all fog is within their ODD. As Mercedes and Nuro did, they could have easily said "light to medium fog." Omitting the intensity of the fog their system can reliably handle comes across as an exaggeration of capabilities.

 

[diplomat33]

I didn't say all fog. I said fog of that intensity. I don't see any difference between fog of that intensity in April 2023, and fog of the same intensity if it were to occur again.

The facts of the matter are that in April 2023, the fog was dense enough that every Waymo driving in it stopped. So we know that at that point in time, fog of that intensity was not within Waymo's ODD.

They put out a blog in August where they say "our Driver is now equipped to handle even heavier levels of fog than before." But it says "heavier fog" not "heavy fog."

I still think it's a problem that Waymo is claiming all fog is within their ODD. As Mercedes and Nuro did, they could have easily said "light to medium fog." Omitting the intensity of the fog their system can reliably handle comes across as an exaggeration of capabilities.

The video where the Waymos stopped due to fog was in April 2023. In August 2023, they said that their system could handle heavier fog. Now, in Jan 2024, they say that fog is in their ODD. So it is entirely possible that there is a progression where they could not do fog in April, they could do heavier fog in August, and now in Jan, they can do all fog, hence why fog is now in their ODD. So I don't think we can assume that they are being dishonest about their ODD. They are only being dishonest about their ODD if they can't handle fog NOW. But maybe now, they can handle all fog.

 

[kabin]

Humans tend to slow for weather conditions and so a valid L4 response should be stopping until conditions improve to tolerable limits. And of course, a more sophisticated sensor suite like Waymo's and Mobileye's will be able to carry on before a pure vision type system. The only other thing to worry about is humans driving faster than conditions warrant but that tends to be the baseline anyway.

 

[Bitdepth]

I didn't say all fog. I said fog of that intensity. I don't see any difference between fog of that intensity in April 2023, and fog of the same intensity if it were to occur again.

The facts of the matter are that in April 2023, the fog was dense enough that every Waymo driving in it stopped. So we know that at that point in time, fog of that intensity was not within Waymo's ODD.

They put out a blog in August where they say "our Driver is now equipped to handle even heavier levels of fog than before." But it says "heavier fog" not "heavy fog."

The video where the Waymos stopped due to fog was in April 2023. In August 2023, they said that their system could handle heavier fog. Now, in Jan 2024, they say that fog is in their ODD. So it is entirely possible that there is a progression where they could not do fog in April, they could do heavier fog in August, and now in Jan, they can do all fog, hence why fog is now in their ODD. So I don't think we can assume that they are being dishonest about their ODD. They are only being dishonest about their ODD if they can't handle fog NOW. But maybe now, they can handle all fog.

I think both of you are saying the same thing but in different ways. Waymo can handle inclement weather in general as part of their ODD hence they have not specified, low, mid and high. There was a time when rain was not part of Waymo's ODD. If it starts to drizzle just slightly, the car will find a place to pull over. Then they added rain into their ODD and we've seen Waymo tackle heavy rain including slightly flooded roads at night even though they have not specified flooded roads as part of their ODD. That's not to say it is perfect, but rain is part of their ODD as is fog because they designed it to handle rain and fog of all types, hence, they have not specified heavy, mid or low they just say inclement weather. Just like how they can pull over for a heavy dense fog they can also pull over for a severe rainstorm if it is deemed unsafe to proceed by the system. A L4 system has to know when it is failing or incapable of handling a situation and enter a minimal risk condition.

I still think it's a problem that Waymo is claiming all fog is within their ODD. As Mercedes and Nuro did, they could have easily said "light to medium fog." Omitting the intensity of the fog their system can reliably handle comes across as an exaggeration of capabilities.

It's not a "problem". Waymo has been very careful with how they deploy their technology which has led many here to call them slow even though by all measure they are well ahead of everyone in this field. They aren't exaggerating the capabilities of their system like how Cruise was. But I do understand the need for clarity and specificity in terms of what the Waymo driver can or can't handle. They've been designing their cars to handle low visibility scenarios including Arizona dust storms.

 

[diplomat33]

Good blog from Kevin Chen on why autonomous trucking on highways is harder that autonomous ride-hailing in city.

Why autonomous trucking is harder than autonomous rideshare

Freeways were supposed to be easy. What happened?

kevinchen.co

Highlights

He proposes the following standard for driverless:

• No driver in the vehicle.
• No guarantee of a timely response from remote operators or backend services.
• Therefore, all safety-critical decisions must be made by the onboard computer alone.
• Under these constraints, the system still meets or exceeds human safety level. This means that on surface streets, the system on its own is capable of driving at least 100k miles without property damage and 40M miles without fatality.

That is a high bar.

Autonomous trucking presents some unique challenges, different from passenger cars:

• Stopping distance for trucks is less than passenger cars. Current sensors have plenty of range for city driving but only a bit more range than the stopping distance of trucks, leaving little margin for error for trucks to avoid a collision.
• Steering and braking on trucks is more difficult since they are bigger, heavier, less agile, and have articulated trailers attached to them rather than being a single body.
• Achieving a minimum risk condition on highways is a lot harder since it is often dangerous to stop in the lane and there is not always a safe spot to pull over.
• Understanding a scene like a construction zone or detour is easier in cities than on highways. An autonomous truck would need to understand the scene and determine correct path in real-time.
• If system uses HD maps and the maps are wrong, the autonomous truck would need to update maps in real-time.
• For cars on city streets, you can somewhat get away with punting unsolved edge cases to remote assistance. You cannot do that with trucks on highways.
• Hardware or software failures could be catastrophic for trucks at highway speeds so you need to eliminate them. You cannot rely on remote assistance to bail the truck out.
• Edge cases happen less often on highways so you need more time and cost to find and train your system.

Kevin believes that these challenges will eventually be solved but he believes we will see robotaxis on highways before we see driverless trucks on highways since cars on highways is an easier problem than trucks on highways.

 

[nativewolf]

That's an interesting viewpoint but I think you'll see Aurura driverless trucks this year, with no safety driver, going from depot to depot in TX. I occasionally meet some of the team, they seem upbeat on the truck side of it.

 

[diplomat33]

According to a recent report from CNBC, which cites two anonymous sources familiar with the matter, GM is poised to end the Ultra Cruise program, with one source indicating that the automaker will refocus its attention on developing Super Cruise, rather than offer two different semi-autonomous driver assist systems. Ultra Cruise was originally slated for launch for the 2023 calendar year.

https://gmauthority.com/blog/2024/01/gm-may-be-canceling-ultra-cruise/