Autonomous Car Progress

[diplomat33]

The faster Tesla progresses, the faster the other fsd developers' houses of cards start to crumble. It'll be a sad day when we all start to realize they were just mostly hollow marketing companies; no offense to them though, as FSD is such a difficult problem, and they needed to keep the hype going to fund their development.

Tesla is the one with L2 that needs constant driver supervision. Waymo and others are the ones with actual driverless cars!

You have it backwards, Tesla is a lot of hype and marketing from the beginning (see 2016 video) and Elon needs to keep it going to fund development.

Think about it: who is hyping robotaxis and has none? Tesla. Who is hyping L5 and only has L2? Tesla. Who has actual robotaxis on public roads today? Waymo.

On the other hand, Waymo, Cruise and others are making real progress and solving real FSD. Waymo went from 1 disengagement per 1,500 miles in 2015 to 1 disengagement per 30,000 miles in 2020. That's progress towards "solving FSD"!

That is a definite trend that shows Waymo's FSD is getting better and better.

Of course, I can be wrong about all of this, but I'll definitely admit my mistake if so

Yes, you are most definitely wrong. You will see as Waymo, Cruise, and others launch robotaxis services to more and more areas in the coming years.

 

[powertoold]

Tesla is the one with L2 that needs constant driver supervision. Waymo and others are the ones with actual driverless cars!

You have it backwards, Tesla is a lot of hype and marketing from the beginning (see 2016 video) and Elon needs to keep it going to fund development.

Think about it: who is hyping robotaxis and has none? Tesla. Who is hyping L5 and only has L2? Tesla. Who has actual robotaxis on public roads today? Waymo.

I agree with you too from a different perspective. I'm looking at this from a rational PoV, and yes, I may be wrong, but like I said, as Tesla progresses quickly, we'll start to see a stark contrast between marketing and reality. It's nice that Tesla is updating fsd beta every two weeks. We'll get frequent reality checks on this.

 

[Daniel in SD]

Even if you assume that, it's still indefensible. You touched on why in your last paragraph, but the short explanation is that not all failures are created equal:

• A once-per 10k mile failure resulting in it failing to see something will occur about once every 100M miles.
• A once-per 10k mile failure resulting in it seeing something that doesn't exist will occur about once every 5k miles.

And since those two cases are likely to be indistiguishable, you have to either trust the worst-case scenario and get phantom braking every 5k miles or trust the best-case scenario and get an accident every 10k miles. So the best-case overall system reliability is still bounded by the single sensor's 10k mile performance at best, or the 5k mile MTBF at worst.

How often does a phantom brake event cause a collision? A collision with injury? A collision with a fatality?
It's the frequency and severity of consequences from the failure that matters.

 

[diplomat33]

I agree with you too from a different perspective. I'm looking at this from a rational PoV, and yes, I may be wrong, but like I said, as Tesla progresses quickly, we'll start to see a stark contrast between marketing and reality. It's nice that Tesla is updating fsd beta every two weeks. We'll get frequent reality checks on this.

I agree time will tell. As FSD Beta gets better, we will certainly see if reality can live up to Elon's hype. I do think that the next 2-3 years will be pivotal for autonomous driving for everybody in general.

And to be clear, the reason I pushed back so hard on your last post is because you seemed to be implying that the AV companies like Waymo, Cruise and others, are all hype and marketing and are not serious about autonomous driving. I think that is going too far. Just watch any conference presentation by Waymo, Cruise or the other dozen AV companies. They all share their work on machine learning to solve various FSD problems. For example, Waymo has shared their ML work on unsupervised optical flow, 3D auto labeling, depth perception with vision, predicting paths up to 8 seconds ahead of multiple objects simultaneously, estimating 3D human pose with both vision and lidar, etc... To act like Tesla is the only company using ML to solve vision or solve FSD is just silly. And all AV companies do a lot of real world testing to see how their autonomous vehicles handle different cases and fine tuning the software to make it better. My point being that you can't reasonably say that Waymo or other AV companies are not taking autonomous driving seriously. They are focused very hard on actually solving the various tasks needed to do autonomous driving reliably and safely.

 

[mark95476]

...Ugh... This is going to really annoy drivers. 30 mph is fast for a cyclist, but a car going that speed max is really going to impede traffic.

I'm glad Waymo more-or-less stopped driving their Pacificas around Mountain View. Sometimes during rush hour they'd be around trying to make lefts, not fully vacating the lane (end sticking out), and wind up blocking traffic.

Cruise was approved to test fully driverless cars (also called Level 4 in industry parlance) in California on October 15th. According to the DMV, Cruise can only test five driverless vehicles “on specified streets within San Francisco.” The vehicles are not allowed to exceed 30 mph, and can’t operate during heavy fog or heavy rain.

Yep, that's a real test. If Tesla operated at 30mph or less and not in rain, I bet it would work just as well if not better.

But 30mph? LOL!!!!

 

[diplomat33]

Cruise was approved to test fully driverless cars (also called Level 4 in industry parlance) in California on October 15th. According to the DMV, Cruise can only test five driverless vehicles “on specified streets within San Francisco.” The vehicles are not allowed to exceed 30 mph, and can’t operate during heavy fog or heavy rain.

Yep, that's a real test. If Tesla operated at 30mph or less and not in rain, I bet it would work just as well if not better.

But 30mph? LOL!!!!

...Ugh... This is going to really annoy drivers. 30 mph is fast for a cyclist, but a car going that speed max is really going to impede traffic.

I'm glad Waymo more-or-less stopped driving their Pacificas around Mountain View. Sometimes during rush hour they'd be around trying to make lefts, not fully vacating the lane (end sticking out), and wind up blocking traffic.

It should be noted that back in 2018, Waymo got a similar permit to test driverless cars in the Mountain View area and their permit included speeds up to 65 mph, rural roads, city streets and highways, and light rain and fog:

Waymo’s permit includes day and night testing on city streets, rural roads, and highways with posted speed limits of up to 65 mph. “Our vehicles can safely handle fog and light rain, and testing in those conditions is included in our permit,” the company says. “We will gradually begin driverless testing on city streets in a limited territory and, over time, expand the area that we drive in as we gain confidence and experience to expand.”

Waymo gets the green light to test fully driverless cars in California

The company is the first to receive a driverless permit in the state.

www.theverge.com

So Waymo's permit is less restrictive than Cruise's permit. I think it could be a sign that Cruise's FSD is less capable than Waymo's FSD which would make sense since Cruise has less autonomous miles experience than Waymo and has lagged behind Waymo in the disengagement rate.

 

[Daniel in SD]

...Ugh... This is going to really annoy drivers. 30 mph is fast for a cyclist, but a car going that speed max is really going to impede traffic.

I'm glad Waymo more-or-less stopped driving their Pacificas around Mountain View. Sometimes during rush hour they'd be around trying to make lefts, not fully vacating the lane (end sticking out), and wind up blocking traffic.

Cruise is operating in San Francisco not Mountain View. I'm guessing the speed limits there are much lower and they'll just avoid streets with faster speed limits.
In fact the Tenderloin district just lowered the speed limit to 20mph.

Tenderloin Celebrates New Speed Limits

San Francisco Bicycle Coalition

sfbike.org

 

[powertoold]

This is going to really annoy drivers. 30 mph is fast for a cyclist, but a car going that speed max is really going to impede traffic.

I just want to point out that the DMV document says that Cruise is limited to certain streets with a speed limit of 30mph. I don't think this means the Cruise car is limited to 30mph.

Most streets in SF actually have a speed limit of 30mph or below (I drive there all the time, including today), so the issue here isn't that it's limited to 30mph streets but that it's limited to certain streets. We don't know which streets or how many streets it's limited to.

Despite all this, I see way'mo Waymo gen5 cars in SF. I rarely see any Cruise cars nowadays.

 

[mark95476]

Thanks for the info.

These autonomous cars will obey the speed limit whether its 30mph or higher/lower, but that'll impede drivers that are driving faster. I try to avoid rush hour traffic so when I'm driving the speed limit is generally the minimum speed.

Aside: Some drivers are absolutely flying (and racing) in Silicon Valley.

I just want to point out that the DMV document says that Cruise is limited to certain streets with a speed limit of 30mph. I don't think this means the Cruise car is limited to 30mph.

Most streets in SF actually have a speed limit of 30mph or below (I drive there all the time, including today), so the issue here isn't that it's limited to 30mph streets but that it's limited to certain streets. We don't know which streets or how many streets it's limited to.

Cruise is operating in San Francisco not Mountain View. I'm guessing the speed limits there are much lower and they'll just avoid streets with faster speed limits.
In fact the Tenderloin district just lowered the speed limit to 20mph.

Tenderloin Celebrates New Speed Limits

San Francisco Bicycle Coalition

sfbike.org

 

[diplomat33]

Despite all this, I see way'mo Waymo gen5 cars in SF. I rarely see any Cruise cars nowadays.

Yep, this guy also says he sees a lot of Waymo 5th Gen around SF.

https://twitter.com/x/status/1424262852475461633

We also have drone footage of the SF Waymo warehouse that shows a bunch of 5th Gen cars.

We also got a couple spy shots of the 5th Gen driving around SF:

https://twitter.com/x/status/1407597036602527746

 

[diplomat33]

I like these little clips that Cruise puts out. They show some city scenarios that AVs have to be able to handle safely and smoothly.

This one shows the Cruise AV navigating around multiple cars with open doors:

https://twitter.com/x/status/1426242939508826116

 

[rxlawdude]

I like these little clips that Cruise puts out. They show some city scenarios that AVs have to be able to handle safely and smoothly.

This one shows the Cruise AV navigating around multiple cars with open doors:

https://twitter.com/x/status/1426242939508826116

Where's the video of an unprotected left turn from a small residential street onto a very busy major divided highway?

 

[Daniel in SD]

Where's the video of an unprotected left turn from a small residential street onto a very busy major divided highway?

I don't think it's possible to navigate San Francisco without making unprotected left turns. Highway 1 and 101 go through San Francisco but I bet all the intersections have lights (still unprotected lefts but I think you're talking about a left without a light).

 

[diplomat33]

Waymo CEO just provided a pretty substantial update on their SF testing IMO:

Waymo is doing 100,000+ autonomous miles per week in SF with EVs (I-Pace).

https://twitter.com/x/status/1426339194767507459

The blog from Feb 2021 has some additional updates:

Specific improvements to the FSD hardware & software:

Throughout that time, we’ve also been gearing our technology to drive in dense cities such as San Francisco. To give just a few examples:

• We’ve optimized the Waymo Driver’s 360° vision system and lidar to navigate the complexities of urban driving. Our highly sensitive cameras can spot traffic lights changing at a long distance – even among the papel picado on 24th Street – to enable smooth driving. And our cameras and lidar can instantly spot a pedestrian sprinting across our path and act appropriately – even when they emerge suddenly from behind a vehicle in the oncoming lane.
• We’ve also designed our software to reason about the context, which is essential for driving safely in busy cities. Our perception system lets our Driver know how to handle a pedestrian, a tree – and a pedestrian carrying a Christmas tree. If we pull up next to a bus by a crosswalk on Beach Street in Fisherman’s Wharf, our Driver can reason that hidden passengers may be getting off, and that they may soon cross the street.
• We’re also building greater flexibility into our driving software to handle unexpected changes to the road. If we’re driving on 19th Avenue during road work and our sensors spot traffic cones and road work signs, our perception system understands that they are guiding us out of the usual lane, and our planning and routing systems can automatically update the vehicle’s route to navigate the new layout.

Waymo has started limited rider testing in SF with Waymo employees, a clear step IMO towards launching a public ride-hailing service in SF:

Well before we deploy a commercial service, our team provides feedback on the product experience, allowing us to validate the progress we’re making and continue improving the rider experience. We’ve now started limited rider testing in San Francisco with Waymo employee volunteers to gather feedback and continue to improve our technology. We’re conducting this testing with enhanced COVID-19 protocols to ensure the safety of everyone involved.

IMO, looks like Waymo is really ramping up their autonomous testing in SF to launch a driverless ride-hailing service soon.

 

[powertoold]

Waymo is doing 100,000 autonomous miles per week in SF with EVs (I-Pace).

Where in the tweet or docs does it say 100,000 autonomous miles per week?

 

[diplomat33]

Where in the tweet or docs does it say 100,000 autonomous miles per week?

It says it right in the tweet:

We’re driving 100,000+ miles per week in SF w/ EVs as we continue to ramp up testing

 

[powertoold]

It says it right in the tweet:

You do realize the blog article you posted is from February 2021?

Also, I still don't see the word autonomous.

 

[diplomat33]

You do realize the blog article you posted is from February 2021?

Also, I still don't see the word autonomous.

The fact that it is testing means autonomous. The autonomous part is clearly implied. They are not driving 100,000 miles manually to test autonomous driving. LOL.

Yes, the blog is from Feb but the tweet from the CEO is from today.

 

[Daniel in SD]

You do realize the blog article you posted is from February 2021?

Also, I still don't see the word autonomous.

Yeah they could be manually driving 100,000 miles a week to map the ~1000 miles of road in San Francisco...

 

[diplomat33]

Yeah they could be manually driving 100,000 miles a week to map the ~1000 miles of road in San Francisco...

No. The 100,000 miles per week is not mapping. You don't need that many miles to map SF. 100,000 miles per week is autonomous testing.