Autonomous Car Progress

[diplomat33]

Seems like that first overtake was from a human-initiated pathing change, just based on how abruptly the path changed, unlike all the other pathing changes

Maybe. Hard to say. Cruise does use remote operators like Waymo who don't control the car but can give the car suggestions. But the car remains in autonomous mode the whole time.

 

[Bladerskb]

Care to revisit your track record?

Four Upcoming Self Driving Level 3 Cars by 2019

New Audi A8 due by end of 2017. Complete autonomous driving for a single lane on the highway while under 35mph. It is based on their jack prototype and uses mobileye eyeq3 chip on a board called zFAS. Its latest sensors include 12 Ultrasonics, 8 Radars, 4 Top View Camera, 1 Wide View Front...

teslamotorsclub.com

Like I said, that was a news aggregate thread which I often post to share compiled information that I attained as a result of my research.
Anyone who is actually curious about my track record needs to see this post below. This is what my opinion and prediction looks like.
It was updated months after that (I believe) to completely eliminate even more companies. Showing you that I wasn't dogmatic.

This is quite different from the "Game over" comments and "Tesla has an immense lead" that you see from others.

@stopcrazypp
I do absolutely believe that to understand others you have to look at their past predictions unless they are no longer making the same predictions with a new date.

You don't see me bringing up anything that @diplomat33 has said because he's not saying the same thing.
I bring up what @powertoold says because he said its game over and made new predictions that just extended the date. Although his recent comments has become a-lot less fantasy.

MobilEye's approach

GM for example hired a company to do map the entire highway system (160,000 miles) with lidar. Marketing stunt. Will be useless unless the lidar-cars keep driving for ever and ever, everywhere. I can't see how they can get away without vision REM / fleet learning, which is cheap/free and...

teslamotorsclub.com

Without further ado here's a post from May 26th 2017:

Keywords: "Everyone is saying 2020 because others are too, you don't wanna say 2025 if others say 2020 cause then say bye bye to your investors, stocks, etc. The whole 2019/2020/2021 dates are just PR announcement. All these companies know that only one or two players will actually have any kind of L4 car in 2020. So when everyone fails, they can just blend in and say "see no one has it too"."

“Toyota’s main objective is safety, so it will not be developing a driverless car,” Seigo Kuzumaki, Toyota’s deputy chief safety technology officer, said during a conference.

Toyota—of All Companies—Defends Drivers, Says It Won’t Build a Fully Autonomous Car

Toyota is definition of clueless, they are not skipping L2 or L3, they don't have L2 or L3.

Now they are saying 2020 for L4, lol give me a break.



Neither Ford nor Toyota will have anykind of L4 in 2020 or 2021. You have to understand we are roughly about 2 from 2020.
2020 is not some long time away. Remember that cars that go into production take about 2 years of testing. A L4 car would need redundancy not only in sensors but in computer system, steering and braking. That entire system must be ready to be pre-production tested in 2018 to even make a 2020 date. This include the platform the system will run on, the exact playment of sensors, and the exact number of sensors. Toyota is still playing russian roullette with their sensors. They are not even close. They are probably 10 years away from L3 even.

This same thing applies to software. Look at the disnegagement from CA DMV. You have to have some kind of progress to even consider 2020/2021. GM for example has cruise which as of last month of 2016 are about 1 in 300 miles and rapidly increasing. Will probably be at a 1 in 1,000-5,000 by the end of the year give or take. Google ofcourse is at 1 in 5,000. Will probably eclipse 10k by years end. My point is, anyone who is trying to deploy something at 2020/2021 must have some progress right now.

There's no magic button that toyota will push and be L4 in 2 years.

You need the software and hardware mostly done now. The only people with the hardware are GM Cruise, Google Waymo, Volvo Drive Me and Audi.




Tesla doesn't have the redundancy and sensors for any kind of L4.
Tesla FSD development began late 2016, anyone who believes Elon will release a L5 that's better than humans with roughly only 1 years of full software development (2018) after their failure to match AP1 parity in 9 months is delusional.

Nio will use mobileye but there is still alot of software/testing that must be written plus they don't even have a car yet.
Lucid will use mobileye but doesn't even have a car yet.
BMW will use mobileye, but will fall victim to the same thing, there's just not enough time to hit 2020/2021, you need your platform ready now.

As the car industry calls it, "pencils down". You need to be pencils down by the end of 2017. You still can't be driving research cars with 20-30 sensors sticking out and aiming to release something in 2 years.

Ford..no

Mercedez is clueless, check their CA DMV report

Baidu already sorta gave in

nuTonomy...nope

Uber...lol they are at 1 disengagement per mile



True.

Delphi for example who says they will have a system ready for OEM in 2019 is still struggling according to their CA DMV report.
Same goes for Bocsh who is claiming L4 in 2020.

Everyone is saying 2020 because others are too, you don't wanna say 2025 if others say 2020 cause then say bye bye to your investors, stocks, etc.

The whole 2019/2020/2021 dates are just PR announcement. All these companies know that only one or two players will actually have any kind of L4 car in 2020. So when everyone fails, they can just blend in and say "see no one has it too".

Google, Audi, Volvo and GM are the one who will have any kind of L4 system in 2020.

When you look at these four systems, you see that everything they have in both software and hardware is mature. They have redundancy in sensors, steering, brakes and they have it in a production-like car (besides GM cruise, although their car looks very production like). They also have mature/maturing software.

 

[mikes_fsd]

Yeah but if you have eyes and can read brail you can read three things at once, thusly you triple the them hours you can devote to reading.

How do you come up with this stuff?

 

[diplomat33]

This might have been posted here already, but maybe not:

Cruise acquires Voyage in another autonomous vehicle merger

It is the latest in a rash of autonomous vehicle mergers.

www.theverge.com

I wonder how an acquisition like this works in practice. Does Cruise just dig through Voyage’s code, throwing everything out except a few good morsels they want to keep?

I think it is more than that. Depending on the terms of the merger, I imagine Cruise could own everything from Voyage, not just the code. So all the facilities and the vehicles could belong to Cruise. And all the Voyage employees could become Cruise employees. I believe Oliver Cameron, the Voyage CEO, is now working for Cruise.

 

[Microterf]

I think it is more than that. Depending on the terms of the merger, I imagine Cruise could own everything from Voyage, not just the code. So all the facilities and the vehicles could belong to Cruise. And all the Voyage employees could become Cruise employees. I believe Oliver Cameron, the Voyage CEO, is now working for Cruise.

I think he was more referring to how does the integration work. What does Voyage have that cruise doesn't have but needs?

 

[diplomat33]

I think he was more referring to how does the integration work. What does Voyage have that cruise doesn't have but needs?

Ok. Thanks. I imagine that Voyage may have human talent that Cruise wanted.

 

[Doggydogworld]

Ok. Thanks. I imagine that Voyage may have human talent that Cruise wanted.

AKA an aquihire. Add a bunch of employees in one shot and curry favor with VCs who got stuck in a loser. Key employees usually have to sign for a couple years, otherwise the deal dies.

Seems like that first overtake was from a human-initiated pathing change, just based on how abruptly the path changed, unlike all the other pathing changes

I think both were human-directed. The first had a lag while the remote monitor assesssed the scene, no lag for the second as he was already engaged.

Cruise leans heavily on remote monitors and sets up tools and procedures assuming they'll be engaged often. That's one reason they max out at 25 mph (for now). Waymo tries to make remote monitors superfluous.

 

[Bladerskb]

I think both were human-directed. The first had a lag while the remote monitor assesssed the scene, no lag for the second as he was already engaged.

Actually the steering wheel icon changes color when remote help is needed. I'm assuming they are continuing displaying it like that.

3m 11s

 

[diplomat33]

Actually the steering wheel icon changes color when remote help is needed. I'm assuming they are continuing displaying it like that.

3m 11s

Thanks. Since the wheel did not change color, it would seem to confirm that there was no remote help. That makes the video even more impressive IMO.

 

[powertoold]

Thanks. Since the wheel did not change color, it would seem to confirm that there was no remote help. That makes the video even more impressive IMO.

I wouldn’t go that far. It’s possible that the steering wheel changes color only when a new path needs to be drawn. There may be different options for path changes. The software might offer different path options, and the remote person can simply choose one without drawing a new one.

in the video where the steering wheel changed color, the guy said that the remote operator made a few clicks to draw a new path.

It’s also possible that Cruise changed their UI to eliminate the color change. We don’t know unless someone has a better description from Cruise.

 

[diplomat33]

I wouldn’t go that far. It’s possible that the steering wheel changes color only when a new path needs to be drawn. There may be different options for path changes. The software might offer different path options, and the remote person can simply choose one without drawing a new one.

That is pure speculation. We don't know that. For all we know, the wheel only changes color on the screen when any remote help occurs. IMO, that would be a more logical assumption than your idea that the wheel only changes for new paths.

 

[powertoold]

That is pure speculation. We don't know that. For all we know, the wheel only changes color on the screen when any remote help occurs. IMO, that would be a more logical assumption than your idea that the wheel only changes for new paths.

Yup, I’m just pointing out that it’s all speculation. We actually don’t know what the steering wheel color means either. Voigt doesn’t mention it during that clip.

 

[diplomat33]

Clips from a 5 hour autonomous drive in Shanghai with zero disengagements from DiDi, with commentary:

 

[stopcrazypp]

Clips from a 5 hour autonomous drive in Shanghai with zero disengagements from DiDi, with commentary:

Probably helps to give some context that DiDi is like China's Uber (in fact they acquired Uber China).

 

[Daniel in SD]

The HD map controversy is easily resolvable. Everyone who claims to use HD maps uses maps that are more detailed than Tesla's. Even as Tesla adds more and more information to their maps it is likely that they will still have the least detailed maps. Therefore the dividing line between regular maps and HD maps can simply be moved so that Tesla never uses HD maps. Problem solved.

Turns out that Tesla does use HD maps. At least this controversy can finally be put to bed.

PlainSite :: Documents :: California DMV Tesla Robo-Taxi / FSD Notes Follow-Up

PlainSite

www.plainsite.org

They partner with TomTom but it's not clear if they buy HD maps from TomTom (HD map | TomTom) though TomTom claims their HD maps are used in 3 million vehicles so it seems plausible.

Other interesting new information is that 88k Teslas have FSD (24k in CA).

 

[gearchruncher]

Turns out that Tesla does use HD maps. At least this controversy can finally be put to bed.

"We briefly barked up the tree of high precision lane line [maps], but decided it wasn't a good idea." -- Elon Musk
"High precision maps and lanes are a really bad idea ... any change and it can't adapt." -- Elon Musk

And for those saying these are not "precision" maps, they are good enough to know where a stop light should be:

Interesting that a human could see the lights, but Tesla's cameras could not.

Other interesting new information is that 88k+ Teslas have FSD (22k+ in CA).

Not quite... They have the HW to "enable the vehicle to use the feature" (HW3) but made it clear not all have purchased the feature. However, these numbers don't make sense, they clearly have more than 88k HW3 cars in the USA.

 

[Daniel in SD]

"We briefly barked up the tree of high precision lane line [maps], but decided it wasn't a good idea." -- Elon Musk
"High precision maps and lanes are a really bad idea ... any change and it can't adapt." -- Elon Musk

And for those saying these are not "precision" maps, they are good enough to know where a stop light should be:
View attachment 668641

Interesting that a human could see the lights, but Tesla's cameras could not.


Not quite... They have the HW to "enable the vehicle to use the feature" (HW3) but made it clear not all have purchased the feature.
View attachment 668640

Tesla sold 38,580 Model 3 in CA alone in 2020 so they must be talking about actual FSD sales not HW3.

Yeah, I know this won't actually stop this silly argument. HD map skeptics will probably always be able to say that Tesla's HD maps are less detailed than those used by other AV companies and therefore superior.

 

[gearchruncher]

Tesla sold 38,580 Model 3 in CA alone in 2020 so they must be talking about actual FSD sales not HW3.

I agree, it's just in direct conflict with "Tesla emphasized that not all these owners have purchased or activated this feature". Tesla's incomplete, conflicting information gets us again!

The really interesting thing is this doc mainly covers just the traffic control detection/response. They need all of this just to know where a traffic light is, one of the most obvious things humans put on roads, and they clearly were only detecting this with vision, so this isn't changing anytime soon.

We also get this!

 

[Daniel in SD]

I agree, it's just in direct conflict with "Tesla emphasized that not all these owners have purchased or activated this feature". Tesla's incomplete, conflicting information gets us again!

The really interesting thing is this doc mainly covers just the traffic control detection/response. They need all of this just to know where a traffic light is, one of the most obvious things humans put on roads, and they clearly were only detecting this with vision, so this isn't changing anytime soon.

We also get this!

Remember that this not a transcript but someone's account of the conversation.
Obviously knowing where to look for traffic lights increases the reliability of the feature which is why HD maps are actually good despite what all the naysayers around here say. Elon's tweets do not match engineering reality (Though it turns out that's not a quote from CJ Moore but the interpretation of the DMV employee on the call).

 

[diplomat33]

Turns out that Tesla does use HD maps. At least this controversy can finally be put to bed.

PlainSite :: Documents :: California DMV Tesla Robo-Taxi / FSD Notes Follow-Up

PlainSite

www.plainsite.org

They partner with TomTom but it's not clear if they buy HD maps from TomTom (HD map | TomTom) though TomTom claims their HD maps are used in 3 million vehicles so it seems plausible.

Other interesting new information is that 88k Teslas have FSD (24k in CA).

"We briefly barked up the tree of high precision lane line [maps], but decided it wasn't a good idea." -- Elon Musk
"High precision maps and lanes are a really bad idea ... any change and it can't adapt." -- Elon Musk

And for those saying these are not "precision" maps, they are good enough to know where a stop light should be:
View attachment 668641

Interesting that a human could see the lights, but Tesla's cameras could not.


Not quite... They have the HW to "enable the vehicle to use the feature" (HW3) but made it clear not all have purchased the feature. However, these numbers don't make sense, they clearly have more than 88k HW3 cars in the USA.
View attachment 668640

Awhile back, I shared this blog about maps:

The Mapping Singularity Is Near

A high falutin’ title to convey how maps are converging and fidelities blending toward a new “medium definition” steady state.

medium.com

There is a middle map in between Standard Definition maps and High Definition maps, called Medium Definition maps (MD maps) which try to be the best of both worlds. They are cheaper and more scalable than HD maps while offering more features and details than standard maps. Technically, I think Tesla Maps would probably be considered MD maps since they have more detail than standard maps for sure but don't have the richness or precision of HD maps.