Autonomous Car Progress

[Bladerskb]

Thats true. But most of the time he is making it more than speed limit.

Though, with Cruise and others we have no idea what they can control remotely ....

Other times he is decreasing it because its taking/approaching a turn too fast, speed is a very huge factor in whether you have a safety disengagement or not.

In other news Amnon just compared Mobileye's supervision camera only system to Tesla FSD and said it will be way over 1,000 hours between MTBF (Safety disengagement to prevent an accident). Him and his interviewer compared it to Tesla FSD current <1 hour MTBF.

That's around 45k miles to 100k miles per safety disengagement compared to Tesla's current < 45 miles per safety disengagement.
Depending on how you calculate the average speed people drive on city streets and what "way above" consists of.

@powertoold i know this will make your blood boil

 

[diplomat33]

Waymo did 2.7M autonomous miles in SF last year!

https://twitter.com/x/status/1486403905290260481

Here are some highlights from the blog:

We evaluated the Waymo Driver’s software performance in increasingly challenging driving situations—day and night, in various weather conditions, from Masonic Avenue’s dense traffic to Valencia Street’s many pedestrians and road closures on a Saturday night.

Last year, we added more vehicles to our fleet of hundreds of Jaguar I-PACEs in San Francisco.

the capabilities of our compute platform that enable the Driver to react quickly in situations like when an emergency vehicle is approaching or someone runs a red light at a busy city intersection

the performance of our sensors in various weather conditions—including San Francisco’s notorious fog, powerful gusts of winds, and sands blowing off Ocean beach

We use a comprehensive set of methodologies to evaluate the performance of our driving software, testing how it completes trips autonomously while avoiding conflicts and adhering to applicable road rules. Dense urban environments present particular challenges against which we rigorously test the Waymo Driver. These include complex intersections, narrow streets, ever changing layouts, and social interactions with other drivers, cyclists and pedestrians. The latter requires the Driver to understand other road users’ intentions and accurately predict their next moves

We’ve engineered the Waymo Driver to drive safely by adapting to the specific situation—like navigating the tight space between a double-parked truck and an oncoming car in the other lane.

we made updates in our fifth generation hardware and perception system that help the Waymo Driver handle foggy conditions even more effectively

According to our recent survey, 94% of our riders in San Francisco are satisfied with the Waymo riding experience and 97% expressed trust in our technology.

With 2.7M miles of testing, I hope Waymo is close to removing the safety driver in SF. I predict they will remove the safety driver in SF this year.

I also think it will be interesting to see Waymo's CA DMV disengagement report. With 2.7M miles, we should get a good enough sample to see how their disengagement rate has improved, especially in light of these improvements they mention in the blog.

 

[diplomat33]

GM CEO Mary Barra got to ride in a driverless Cruise robotaxi.

 

[Doggydogworld]

GM CEO Mary Barra got to ride in a driverless Cruise robotaxi.

"It will reduce congestion"...... somebody hasn't thought things through.

 

[diplomat33]

"It will reduce congestion"...... somebody hasn't thought things through.

I think that is a common assumption actually. The idea is that with ride-sharing, you could eventually replace all the current cars in a city with a smaller number of robotaxis that could operate 24/7 to take people where they need to go.

 

[Daniel in SD]

"It will reduce congestion"...... somebody hasn't thought things through.

Maybe if it allows you to get rid of a bunch of street parking?
Also maybe you could have a bunch of Cruise AVs caravan a few feet apart?
There's also the claim that a bunch of congestion is caused by people looking for parking.
Maybe AVs will encourage people to use public transit more? Take Cruise to BART and then another Cruise to your final destination.

Though maybe she's talking about the Cruise Origin which will be used for shared rides.

 

[Sporty]

I think that is a common assumption actually. The idea is that with ride-sharing, you could eventually replace all the current cars in a city with a smaller number of robotaxis that could operate 24/7 to take people where they need to go.

I’m going to send my Optimus out for groceries and other errands all the time. It’s going to do a lot of traveling by cars.

 

[stopcrazypp]

I think that is a common assumption actually. The idea is that with ride-sharing, you could eventually replace all the current cars in a city with a smaller number of robotaxis that could operate 24/7 to take people where they need to go.

That was the same pitch by the "ride-sharing" companies too. Instead what happened was given the cars drove around to look for fares (or be ready for a fare), it increased the congestion. Also relatively few trips actually were done with the ride shared (many done with only a single group of riders). It also displaced trips that otherwise would have used other modes of transportation (like walking, biking, taking the bus or other mass transit) and its convenience meant some people would make a trip they otherwise wouldn't have made with other forms of transport.

It's certainly very convenient, but reducing congestion isn't really what is happening (at least with the current system, maybe with some kind of advanced scheduling/coordination system it might be possible).

 

[coselectric]

Interesting choice for GM to evangelize their automated vehicle tech in a model that has been pulled from the market because it has a tendency to burn their customers' houses down.

 

[Daniel in SD]

Interesting choice for GM to evangelize their automated vehicle tech in a model that has been pulled from the market because it has a tendency to burn their customers' houses down.

Doesn't it make more sense to use them as robotaxis since then they won't be parking in garages?
Anyway the Cruise Origin is supposed to go into production in 2023.

 

[ThomasD]

How many cars would Cruise need if everyone who had a car in the San Francisco metro area gave them up? Say they ran them 24/7 with a 15 minute wait time

 

[ThomasD]

They will still need maintenance facilities. Be able to charge many cars at once. Commercial insurance. Customer service reps. to handle issues with customers and vehicle issues while the vehicle is in operation.

 

[Daniel in SD]

How many cars would Cruise need if everyone who had a car in the San Francisco metro area gave them up? Say they ran them 24/7 with a 15 minute wait time

You're always posting these math problems aren't you? haha
8 million people in Bay Area (1/5th of state)
340 billion miles per year in CA
So let's say 68 billion miles per year in bay area
Average speed 20mph?
Peak utilization 4x average?

68e9/20*4/(365*24) = 1.5M Cruise vehicles

They claim to have the factory capacity to build "tens of thousands" (Hitting the Accelerator).

Now if they were able charge a dollar per mile (that's less than Uber) that would be $68 billion in revenue per year, more than Tesla's 2021 revenue of $54B.

They will still need maintenance facilities. Be able to charge many cars at once. Commercial insurance. Customer service reps. to handle issues with customers and vehicle issues while the vehicle is in operation.

Yes.

 

[coselectric]

Doesn't it make more sense to use them as robotaxis since then they won't be parking in garages?

Sure, just as long as they are equipped with smoke detectors and emergency stop buttons.

 

[stopcrazypp]

How many cars would Cruise need if everyone who had a car in the San Francisco metro area gave them up? Say they ran them 24/7 with a 15 minute wait time

That seems to be a false comparison though, as many cars owned in SF are for longer trips, trips that would typically be outside of a local robotaxi network. If you include coverage or longer distant trips (like SF to LA, Sacramento, SJ, or even closer to across the Bay), then the fleet size required may be a lot more and no sure if it would work out.

Most of the robotaxi networks so far seem to be for relatively short distance travel.

 

[rbt123]

How many cars would Cruise need if everyone who had a car in the San Francisco metro area gave them up? Say they ran them 24/7 with a 15 minute wait time

Manhattan had 13500 taxis and 1.6% modal share for commute; so each taxi in the city handled 2 commute related customers. 35 trips per day per vehicle, but most were not for commuting purposes.

So for a rough estimate you might take the work-force that currently drives and divide by 2.

There are ways of reducing that using pricing strategies like airlines. Cheaper fares for booking trips in advance and outside peak hours. Last-mile trips with high-capacity transit in the middle also significantly increase capacity; they might do 10 neighbourhood trips to a nearby train station in an hour vs a single end-to-end commute.


A good start would be replacing the existing Taxi and ride-share market, which is about 800,000 vehicles across North America. That'll be a few years production.

 

[Dutchy Ron]

Full self-driving, as a reminder, is Tesla’s name for its autonomous vehicle capabilities. As it stands, a beta version — a suite of driver-assistance features that do not make the vehicle autonomous, where drivers must keep their hands on the steering wheel and be ready to take over at any time — is currently being used in nearly 60,000 U.S. vehicles, per Reuters. But Musk is saying that he expects Tesla to achieve the holy grail that is a truly autonomous vehicle this year.

As Jason Torchinsky eloquently explains at Jalopnik, Musk has now been making this claim for nine years in a row. So the chance that Tesla actually delivers on this promise this year is extremely low.

 

[stopcrazypp]

Full self-driving, as a reminder, is Tesla’s name for its autonomous vehicle capabilities. As it stands, a beta version — a suite of driver-assistance features that do not make the vehicle autonomous, where drivers must keep their hands on the steering wheel and be ready to take over at any time — is currently being used in nearly 60,000 U.S. vehicles, per Reuters. But Musk is saying that he expects Tesla to achieve the holy grail that is a truly autonomous vehicle this year.

Not exactly, this was what he actually said:
"I would be shocked if we do not achieve Full Self-Driving safer than a human this year. I would be shocked".
Given Tesla seems to claim AP is safer than a human already, I'm not sure that's very meaningful, as it's not a very high bar. I remember 2x being thrown out and 10x being the ultimate goal.

Sounds to me more like he expects a general FSD City Street release (AKA "feature complete" FSD, AKA end-to-end L2, but nothing like achieving L3 or L4/L5). That said, 60k is a lot of cars, the previous estimate was around 20k.

As Jason Torchinsky eloquently explains at Jalopnik, Musk has now been making this claim for nine years in a row. So the chance that Tesla actually delivers on this promise this year is extremely low.

 

[Daniel in SD]

Not exactly, this was what he actually said:
"I would be shocked if we do not achieve Full Self-Driving safer than a human this year. I would be shocked".
Given Tesla seems to claim AP is safer than a human already, I'm not sure that's very meaningful, as it's not a very high bar. I remember 2x being thrown out and 10x being the ultimate goal.

Sounds to me more like he expects a general FSD City Street release (AKA "feature complete" FSD, AKA end-to-end L2, but nothing like achieving L3 or L4/L5). That said, 60k is a lot of cars, the previous estimate was around 20k.

So what is he talking about here?

https://twitter.com/x/status/1450868052040507392

Is there anything that Elon could say that would convince you he's talking about robotaxis?

 

[daktari]

Not exactly, this was what he actually said:
"I would be shocked if we do not achieve Full Self-Driving safer than a human this year. I would be shocked".
Given Tesla seems to claim AP is safer than a human already, I'm not sure that's very meaningful, as it's not a very high bar. I remember 2x being thrown out and 10x being the ultimate goal.

Sounds to me more like he expects a general FSD City Street release (AKA "feature complete" FSD, AKA end-to-end L2, but nothing like achieving L3 or L4/L5). That said, 60k is a lot of cars, the previous estimate was around 20k.

No. The question was on Level 4 and he talks about 10.000 times safer than humans. He could try to dodge the question of course but that seems a bit strange way to do it.

"Very small number of interventions pr mile"

Tesla (NASDAQ:TSLA)
Q4 2021 Earnings Call
Jan 26, 2022, 5:30 p.m. ET

Martin Viecha

Thank you. And the last question from investors is, Elon mentioned Level 4 autonomy could be achieved this year. Is it based off initial FSD beta rollout experience, or is Level 4 ability predicated on Dojo being completed online?

Elon Musk -- Chief Executive Officer and Product Architect

As mentioned earlier, Dojo is not required for full self-driving. You know, it should have a positive effect on the cost of training networks. You know, it's not just a question like does it get to full self-driving but really kind of like the "march of nines" of reliability, is it 99.999% reliable or 99.999999% reliable. This is -- it gets nutty.

So, obviously, we want to get to close to perfection as possible. So, frankly, being safe than human is a low standard, not a high standard. People are very, very lossy, often distracted, tired, you know, texting. Anyway, it's remarkable that we don't have more accidents.

So, it's -- yeah. So actually being better than a human, I think, is relatively very forward, frankly, how do you be 1,000% better or 10,000% better. Yeah, that's what, you know, gets much harder. But I think anyone who's been in the FSD beta program, I mean, if they were just to plot the progress of the beta interventions per mile, it's obviously trending to, you know, a very small number of interventions per mile and pace of improvement is fast.

And there's several profound improvements to the FSD stack that are coming, you know, in the next few months. So, yeah, I would be shocked if we do not achieve full self-driving safer than human this year. I would be shocked.