When Tesla will catch up with Waymo and Cruise at delivering L4?

[OncomingStorm]

Hard to believe but apparently some people actually think FSD will get a lot better magically within a year.

I think they have not looked at the current disengagement rate to realize we need 100x to 1000x improvement.

i'm not saying FSD will be level 4 or anything close to it by the end of the year. i think it will be a loooong time before we can take naps in our car while driving lol... however, given the rate of progression over the last year i can totally see some very significant progress being made over the next. Again, we'll still be in the drivers seat, but imho, its very possible you'd have disengagement rates so low that its pretty close to functional self driving, kinda like highway AP now (polishing, improving safety and getting these features through regulatory bodies will take many years).

I only say this because of the way they are crowd sourcing raw data. to have that many end points returning functional data its extremely powerful. and unlike other vendors this is real world data, not trained drivers in designated areas, which means the limits are going to be pushed much farther, further increasing the value of the data. assuming they continue adding users (at some point lol) they could achieve a rate of progress boarding on exponential, if they havn't already since no one really knows lol.

when talking about robo taxi's tho, i 100% think we'll see lots of companies in geofenced cities pop up long before tesla's robotaxi service, however, once they do achieve it they could very well capture the entire market since they will have no boundaries. it would not only encompass taxi services but i think it would also dent airlines and other forms of travel since you could like hail a tesla and go across the country if you wanted. of course that outcome has a lot of assumptions but so does any

 

[EVNow]

however, given the rate of progression over the last year i can totally see some very significant progress being made over the next.

Can you quantify ? My estimate for disengagements is 1 in 10 miles. What was it last year ?

Again, we'll still be in the drivers seat, but imho, its very possible you'd have disengagement rates so low that its pretty close to functional self driving, kinda like highway AP now (polishing, improving safety and getting these features through regulatory bodies will take many years).

So, what kind of disengagement rate will we see by the end of this year ? 1 in 100 miles, 1,000 miles, 10,000 miles (human driver).

 

[Daniel in SD]

Can you quantify ? My estimate for disengagements is 1 in 10 miles. What was it last year ?



So, what kind of disengagement rate will we see by the end of this year ? 1 in 100 miles, 1,000 miles, 10,000 miles (human driver).

It was 0 disengagements a year ago:

It was also between Level 3 and Level 4 a year ago.

 

[t3sl4drvr]

So, what kind of disengagement rate will we see by the end of this year ? 1 in 100 miles, 1,000 miles, 10,000 miles (human driver).

Just as an annoying benchmark. Here are 2021 disengagement rates in California

Observations

• Tesla is not reporting at all
• Waymo is down to 8k from 30k miles in 2020. Apparently due to switching their California fleet of 700 cars fully to Jaguar iPace
• AutoX, Cruise and Waymo (with their old Pacificas) are already is sufficiently widely proven to have > 10,000 miles / disengagement rate. An all of them have license to drive in California without safety drivers.

In any case, Tesla is several years behind the competition for L4.

Interestingly the most driverless miles (no safety driver and thus no interruptions) were reported by Apollo (Baidu): 9565 miles. This puts Volvo's claims for bringing autonomous Ride Pilot feature on market to a perspective (I believe they are using Apollo technology for L4)

 

[EVNow]

Just as an annoying benchmark. Here are 2021 disengagement rates in California
View attachment 772670

I hat-top to all those Lidar + HD Maps AVs with disengagement rates < 50 miles.

Interestingly the most driverless miles (no safety driver and thus no interruptions) were reported by Apollo (Baidu): 9565 miles. This puts Volvo's claims for bringing autonomous Ride Pilot feature on market to a perspective (I believe they are using Apollo technology for L4)

I thought Volvo was using NVidia. Also, Volvo is claiming L3 for Highways ... not city. I'm most of the stats above are for city ?

Wiki - Consumer AV - Status Tracking Thread

This thread is to track the progress of Consumer AVs - from ADAS to full self-driving. I'll keep this info updated. Let me know if you have any updates & corrections.

teslamotorsclub.com

 

[t3sl4drvr]

I thought Volvo was using NVidia. Also, Volvo is claiming L3 for Highways ... not city. I'm most of the stats above are for city ?

I am probably mistaken about Apollo. Volvo seems to be collaborating with many autonomous driving partners (or switching them yearly). Looks like Ride Pilot is actually developed by company called Zenseact. They will provide both L2 mode "Cruise" and L4 mode "Ride" on limited set of roads. Strangely, it seems that they have not been testing in California at all (or maybe they operate under some other name here). In any case, HW is from NVidia.

 

[EVNow]

I am probably mistaken about Apollo. Volvo seems to be collaborating with many autonomous driving partners (or switching them yearly). Looks like Ride Pilot is actually developed by company called Zenseact. They will provide both L2 mode "Cruise" and L4 mode "Ride" on limited set of roads. Strangely, it seems that they have not been testing in California at all (or maybe they operate under some other name here). In any case, HW is from NVidia.

This not testing at all is the important thing.

Apparently some companies think it’s ok to do limited real world testing, a lot of simulation and get the L3/L4 certification. I hope CA asks for a few million miles of testing before allowing them to claim L3 certified.

 

[OncomingStorm]

Can you quantify ? My estimate for disengagements is 1 in 10 miles. What was it last year ?

Sadly i'm not sure anyone can say for sure since they dont publish this data, however we can extrapolate some useful information from anecdotal reports from various sources; forums like this, youtube, reddit, the myriad of other similar sites, personal accounts, personal experience. etc... Arguably not the best data source, youtube is good for one thing, the youtubers tend to do the same routes over and over and while yes the general behavior of the car has had many ups and downs, the actual disengagement rate of those, and other, reports has significantly reduced over the last year or whatever time frame you want to use. however, more importantly, reports of serous safety disengagements has dropped off the map. yes it will still do dumb dangerous things on rare occasion but as a whole most of the disengagements you see reported by people are for much smaller things, like checking a curb or getting to close to something while not actually hitting it or simply freaking people out lol.

I think trying to guess at the disengagement rate is a bit of a false flag, we dont want to guess at stats and then argue about them as that is kinda pointless imho so i'm reluctant to actually put a number on it and prefer to simply look at trends, if that makes sence.

also, i am sure they are focused on safety above comfort, so you kind of have to ignore some things that make you freak out a bit but technically work, this is like proof of concept, if they can get it working reguardless of how it feels, they can always adjust it later to improve comfort. very standard programming approach.

So, what kind of disengagement rate will we see by the end of this year ? 1 in 100 miles, 1,000 miles, 10,000 miles (human driver).

again rather tough to say, even if we had all the available data and could statistically model the trend there a LOT of unknowns lol. that said, gun to my head lol, my guess would be we will be at or above human error rates. now bear in mind, to me, this means fleet wide, so you could absolutely have areas of the country where specific drive styles or conditions that cause FSD performance to fall off the map lol, as is always the case with any large scale system, but there will also be places where the rate is 1 in a billion miles cause there are no edge cases to be encountered and no traffic or something like that.

 

[t3sl4drvr]

Apparently some companies think it’s ok to do limited real world testing, a lot of simulation and get the L3/L4 certification. I hope CA asks for a few million miles of testing before allowing them to claim L3 certified.

There is a certification? I do not think California is doing any certification testing.

Instead there is a permit needed for operating self-driving vehicles in the state. Not sure what requirements, if any the permit application process has, but the permit can be taken away in case of accidents.

 

[EVNow]

There is a certification? I do not think California is doing any certification testing.

Instead there is a permit needed for operating self-driving vehicles in the state. Not sure what requirements, if any the permit application process has, but the permit can be taken away in case of accidents.

Self-driving car regulations don’t cover ADAS. But L3/L4 probably is covered ?

UN has a standard that companies have used to claim L3 certification. It covers highway driving in heavy traffic below 60kmph / 37 mph. I think that is what Volvo is talking about. It would be the first time anyone would be trying it in US.

 

[t3sl4drvr]

Self-driving car regulations don’t cover ADAS. But L3/L4 probably is covered ?

In California, you need to apply for permit "to deploy autonomous vehicles on public streets" with this application. It links to relevant regulations.

Cruise, Nuro, and Waymo does have a permit for deployment (actual customers in cars). In addition to them, Apollo, Autox, Weride, and Zoox has a permit to do testing without a driver in car.

Tesla do not have a permit to do any driverless testing or deployment, but finally has a permit to test with a driver. Now that Tesla did not report any autonomous driving miles in 2021, but is finally doing testing, it is interesting to see how many miles and disengagements they report in 2022.

 

[EVNow]

In California, you need to apply for permit "to deploy autonomous vehicles on public streets" with this application. It links to relevant regulations.

Cruise, Nuro, and Waymo does have a permit for deployment (actual customers in cars). In addition to them, Apollo, Autox, Weride, and Zoox has a permit to do testing without a driver in car.

Yes, we have discussed that application before.

But, does that cover "selling" autonomous cars to customers - that is different from deploying them, I guess.