Autonomous Car Progress

[Bladerskb]

Bold statement. I'd be surprised if Tesla is 5+ years ahead of everyone else. I'm sure Waymo or other driverless companies are going to be in a dozen cities before Tesla has a useful uber like feature. Although that doesn't disagree with your statement about being nation wide. It will be 5 or more years before Tesla reaches uber like FSD. I do think they are ahead, but 5+ years seems too much. Maybe 2+ years.

What i have discovered is some Tesla fans are incapable of using logic and utilizing reasoning when discussing anything Tesla related and are like conspiracy theorists and flat earthers.

When someone says that Tesla FSD is supposedly 5+ years ahead. It showcases that.
You would have to ignore reality like flat earthers. This is why most tesla fans ignore mobileye's supervision, Huawei autopilot or even the recent GM ultra cruise. If Tesla FSD is supposedly 5+ years ahead, then this is the easiest thing to prove factually beyond any reasonable doubt as 5 years is such a huge gap and easily demonstrable.

Let's take GM UltraCruise for example that works practically *everywhere in the US. Ultra Cruise is set to launch starting on GM's EV Lyriq and others GM cars starting in 2023. If Ultra Cruise were to launch on October 20th 2023, Exactly 3 years after FSD Beta initial release to ~80 (mostly) Tesla super fans. Ultra cruise would have to be significantly worst than even the first FSD Beta software to be 5 years behind.

And if you remember how bad the first FSD Beta software was, it couldn't even make basic left/right turns consistently and struggled with basic lane keeping and couldn't handle simple things like yield signs. It was just horrific. I could keep going. People should watch videos of the initial software release to remember how bad it was.

UltraCruise has to be SIGNIFICANTLY and orders of magnitude WORSE than that to be 5+ years behind.

Its simply illogical, makes no sense. But actually it does because most Tesla fans DO NOT use their brain when talking anything tesla related.

Now that's just UltraCruise. Then there is Huawei Autopilot that is releasing in December 2021. Again it would have to be thousands of times worse than the initial FSD beta from Oct 20th 2020 for it to be 5+ years behind. Especially with them operating in an environment that is orders of magnitude harder to drive in than the US.

Then finally you have supervision from Mobileye that works anywhere on the planet that has REM Map. All of US, Canada, Europe, Japan, China, etc.
Supervision launches with an OTA update in 2022 and works all over china. Again just like Huawei, Super Vision has to be thousands of times worse than the initial FSD beta to be 5+ years behind.

It literally makes ZERO sense.

And with GM Ultracruise we can actually test routes and compare them with videos of the initial FSD Beta when Ultracruise launches to see how many years Ultracruise is behind and whether they are behind at all.

But that requires logic and reason and we can't have that.

 

[mark95476]

2) That'll never end and Tesla owners expect regular SW updates. A ~2 week FSDbeta release cycle is unreal.

My main takeaways from 10.2 in a nutshell:

1) Lots of "minor" issues, 90% of which will be addressed within 6 or less months
2) Tesla is still working on massive video ingestion / auto-labeling to improve the BEV geometry predictions
3) Uncertain whether HW3 will be able to handle the inference from millions of video clips

Tesla continuously tweak HW3 firmware, but I bet they're cooking up a bunch of advanced functions and features that won't get backported when HW4 is released. I was somewhat hoping FSD would get pushed to HW4.

Irrespective what actually happens - I don't think people with HW3 will be left out. They will just upgrade them instead of handing money back. Anyway with Tesla probably already making $2B profit a quarter, upgrade cost will be easy to absorb.

I guess that's progress (or just moving goalposts). From never going to see the button and always negging/doubting anything FSD to this...

Bold statement. I'd be surprised if Tesla is 5+ years ahead of everyone else. I'm sure Waymo or other driverless companies are going to be in a dozen cities before Tesla has a useful uber like feature. Although that doesn't disagree with your statement about being nation wide. It will be 5 or more years before Tesla reaches uber like FSD. I do think they are ahead, but 5+ years seems too much. Maybe 2+ years.

 

[Bladerskb]

Running into objects is not a minor issue. It is rule number one, don't crash into objects or people.

And here is what he said about 12 months ago about how easy Tesla is able to get to L5 with about 150k miles per disengagement in 6-9 months. Now 12 months later and Tesla is still on average encountering a safety disengagement every drive. These people are like flat earthers, they are literally clueless and start from a conclusion and try to make things up so it can add up to it.

Sorry, but this is the wrong logic to apply. Disengagement or development improvement doesn't have to be linear. For example, let's say drivers often have to disengage every 10 miles because the car doesn't get into the correct left turn lane. If Tesla fixes that one problem, it's possible drivers will only have to disengage every 100 miles. I think disengagement improvement can be exponential.

Many of the disengagements we've seen are mostly:
Gets into wrong turn lane
Moves into different lane over complicated intersections
Difficulty turning into narrow roads with cars

As for my estimate of 6-9 months, that's unbelievable to me. But based on what we're seeing, it's possible.

 

[Dutchy Ron]

Rain impairs performance of driver-assistance technology - ET Auto

ADAS: Researchers at AAA, a federation of North American motor clubs, found that vehicles' auto emergency braking systems, in several instances during simulated rainfall, no longer recognized stopped vehicles ahead and that vehicles' lane-keeping systems performed considerably worse.

auto.economictimes.indiatimes.com

There is no Tesla FSD that works as an Uber. There were supposed to be 1 million of them in 2020 but there are currently none, not in a geofenced area and most certainly none anywhere on this planet.

Would we be surprised? Do we not have documented examples of some issues Waymo has? We know Waymo does not like to drive in the rain but JJRicks has it on record showing it. Waymo car detects rain, informs JJRicks on the rider screen, decides it wants to pull over, hands control over to the safety "driver" since there is one in the car, and informs remote support who then calls JJricks to report the issue. All recorded on video. Whats not transparent about that?

Rain impairs performance of driver-assistance technology​

Rain impairs performance of driver-assistance technology - ET Auto

ADAS: Researchers at AAA, a federation of North American motor clubs, found that vehicles' auto emergency braking systems, in several instances during simulated rainfall, no longer recognized stopped vehicles ahead and that vehicles' lane-keeping systems performed considerably worse.

auto.economictimes.indiatimes.com

With real time OTA in car HD Maps with Road DNA you dont’ have this problem when it rains. It depends if a car uses camera, sensors or lidar when it’s dark or bad weaher (rain, snow, fog, etc.) Additional services like real time weather service and hazard wanring will contribute to the safety of the driver as well.

I agree that humans don’t need HD Maps but for sure FSD cars will, for safety reasons and rainy days. Or you are fine driving yourself when it rains. That’s why I can imagine Elon will change his mind, approach and platform soon for the new cars and models. In that case he should refund all Tesla owners who paid for the FSD. He used this free money what was paid in advance for new investments, very clever. If Apple joins Tesla this issue is solved ;-)

 

[EVNow]

I guess that's progress (or just moving goalposts). From never going to see the button and always negging/doubting anything FSD to this...

Moving goal posts / adjusting to reality. Atleast they are delivering cars.

INFINITI LE Concept – A Vision of Zero Emission Luxury Revealed at New York International Auto Show

The INFINITI LE Concept, which makes its world debut at the 2012 New York International Auto Show, is more than just the latest in the series of dramatically styled, technologically advanced concepts from INFINITI.

usa.infinitinews.com

"In a rapidly changing world, INFINITI is well ahead of the game."

 

[rxlawdude]

I agree that humans don’t need HD Maps but for sure FSD cars will, for safety reasons and rainy days. Or you are fine driving yourself when it rains. That’s why I can imagine Elon will change his mind, approach and platform soon for the new cars and models. In that case he should refund all Tesla owners who paid for the FSD. He used this free money what was paid in advance for new investments, very clever. If Apple joins Tesla this issue is solved ;-)​

So what happens when there's construction on a mapped road the day after it's mapped? By definition, isn't the map outdated once any change is made?

 

[Dan D.]

Rain impairs performance of driver-assistance technology - ET Auto

ADAS: Researchers at AAA, a federation of North American motor clubs, found that vehicles' auto emergency braking systems, in several instances during simulated rainfall, no longer recognized stopped vehicles ahead and that vehicles' lane-keeping systems performed considerably worse.

auto.economictimes.indiatimes.com

Rain impairs performance of driver-assistance technology​

Rain impairs performance of driver-assistance technology - ET Auto

ADAS: Researchers at AAA, a federation of North American motor clubs, found that vehicles' auto emergency braking systems, in several instances during simulated rainfall, no longer recognized stopped vehicles ahead and that vehicles' lane-keeping systems performed considerably worse.

auto.economictimes.indiatimes.com

With real time OTA in car HD Maps with Road DNA you dont’ have this problem when it rains. It depends if a car uses camera, sensors or lidar when it’s dark or bad weaher (rain, snow, fog, etc.) Additional services like real time weather service and hazard wanring will contribute to the safety of the driver as well.

I agree that humans don’t need HD Maps but for sure FSD cars will, for safety reasons and rainy days. Or you are fine driving yourself when it rains. That’s why I can imagine Elon will change his mind, approach and platform soon for the new cars and models. In that case he should refund all Tesla owners who paid for the FSD. He used this free money what was paid in advance for new investments, very clever. If Apple joins Tesla this issue is solved ;-)

I wish these tests would be a little more daring.

They tested cars at speeds of 35 mph (56 km/h).

Why don't they test faster? On almost no city streets here are people driving that slowly, even during the rain. I want to know how effective these systems are in real-world driving as well as their "everybody drives the speed limit" fantasies.

 

[svusa]

2) That'll never end and Tesla owners expect regular SW updates. A ~2 week FSDbeta release cycle is unreal.



Tesla continuously tweak HW3 firmware, but I bet they're cooking up a bunch of advanced functions and features that won't get backported when HW4 is released. I was somewhat hoping FSD would get pushed to HW4.



I guess that's progress (or just moving goalposts). From never going to see the button and always negging/doubting anything FSD to this...

Is Tesla really ahead or it is illusion that they are ahead? I have been driving M3 since 2018 and while UI and lane keeping has improved.. It still makes different decisions under similar type of situations. That introduces elements of uncertainty and as a driver you have no idea what AI is thinking and you will have to take over to avoid getting into accident.

Looking at 10.2 FSD beta videos.. I feel that similar issues exist and system can’t make decisions 100% of the time with certainty.

I think tech is not ready yet.. may be 5-10 years it will look different with newer tech/processing power etc.

In meanwhile, we will likely to see L4 Robotaxis by other companies in cities around the world. It is also likely that Chinese Tech will crack FSD before US.

Just my two cents…

 

[Dutchy Ron]

So what happens when there's construction on a mapped road the day after it's mapped? By definition, isn't the map outdated once any change is made?

To be clear, Tesla should add HD Maps for more safety, FSD is not possible only with HD Maps, it’s the combination with camera, sensors and lidar (incl HD Maps) that give most safety for the driver in all weather conditions.

 

[Dutchy Ron]

So what happens when there's construction on a mapped road the day after it's mapped? By definition, isn't the map outdated once any change is made?

HD Maps can be updated real time when the mapmaker receives all the data from the OEM’s and local governments, TomTom for example, receives a lot of probs (data) from cars (sensors, camera and lidar) and can update their HD maps 95% fully automated. If a car is connected and can receive OTA real time HD Maps it has always actual HD Maps. So HD Maps needs to be used in combination with sensors, lidar and camara’s. With HD Maps you can look ‘around the corner’ and localize your car on the map (lanes) with 10cm acurancy. If HD Maps are needed or not is discussed a lot. At the time Elon launched the first Tesla there were no global HD Maps. To make and maintain HD Maps is (was) very expensive because they were made by the mapmaker with their own cars. So in that time Elon has to find another approach and that was very innovative for that time.
But now (5 years later) a lot of cars having sensors, lidar and camera’s, so there’s a lot more data available and it’s much more easier to update HD Maps. There are not many companies who can manage all this data and are able to update their global maps almost fully automated. In my opinion the top 5 companies are TomTom, Baidu, Navinfo, HERE and TeleNav. Microsoft Azure Maps used in the Connected Vehicle Platform (MCVP) are from TomTom, they will be used by VW next year and also by Nissan.
I do not want to start a new discussing about the question if HD Maps are needed (required) for safety reasons. I only think that Elon has a problem with the law of the handicap of a head start, the first mover disadvantage. Tesla will stuck at level 3 in my opinion if Elon is not willing to redesign his vision. Just my opinion ;-)
Next year we will see more use of HD maps in L3 EV/AV cars, also in China. They will prove that the driver experience is much better than from Tesla. Don’t get me wrong I have a lot of respect for Elon, he’s a real visionair and innovator but I think he also realizes that he’s drving into a dead end road with the FSD promise in the existing Tesla’s.

So to anwer the question, HD Maps are not only updated anymore by cars from mapmakers like 5 years ago. Mapmakers are using lots of data from all OEM’s. So HD Maps are getting cheaper to maintain and you only need the HD Map for the area where you are driving. You don‘t need gigs of data in your car of all HD Maps of the whole world. Connected to 5G will make it a lot more easy. Elon said HD Maps are expensive and not scalable. He was right 5 years ago not anymore.

 

[Dutchy Ron]

WeRide Introduces Robovan, the First Level 4 Autonomous Cargo Van to Operate in China​

The technology developed by the Chinese company includes in-house built HD maps that cover thousands of kilometers through various cities, plus real-time localization, thanks to a series of integrated sensors. The result is a Level 4 autonomous vehicle that can drive even through tunnels that are longer than 0.6 miles (1 km), in challenging weather conditions, such as heavy rain. The future cargo van will be able to provide delivery services both day and night, in any kind of weather.

WeRide Introduces Robovan, the First Level 4 Autonomous Cargo Van to Operate in China

“Smart urban logistics” is not just a catchy phrase, but a vision of what cargo transportation will look like in the near future. WeRide, the Chinese autonomous driving company, is exploring this area also, after its successful robotaxi operations, with a brand-new vehicle: the robovan.

www.autoevolution.com

 

[powertoold]

HD maps are stupid. Companies use them because they have to. They don’t have an alternative. They don’t believe it’s possible for cars to drive like humans do.

 

[diplomat33]

HD maps are stupid. Companies use them because they have to. They don’t have an alternative. They don’t believe it’s possible for cars to drive like humans do.

This is insanely wrong and ignorant. HD maps are not stupid. They increase safety and provide useful semantic info. Everybody who works in the AV industry understands their value. That's why they use them.

 

[powertoold]

This is insanely wrong and ignorant. You sound like a total idiot when you post things like this.

HD maps are not stupid. They increase safety and provide useful semantic info. Everybody who works in the AV industry understands their value. That's why they use them.

And AVs don't use HD maps because they can't drive the car with just perception. They use HD maps WITH Perception. They use both together to give the car increased reliability in its perception so that it can make better driving decisions.

What's becoming evident is that HD maps are leading to a local maximum. It's been how many years since they were first used?

HD maps are very tempting to use, for the very reasons we see 10.2 fail. But, Elon and everyone keeps repeating the same thing, which is that vision needs to be "solved" before FSD can be achieved. Basically, vision is the rate-limiting factor for FSD development, not HD maps. Once vision is adequate for multiples of human safety, then HD maps aren't useful, they'd just be more noise.

 

[DanCar]

HD maps are stupid. Companies use them because they have to. They don’t have an alternative. They don’t believe it’s possible for cars to drive like humans do.

Humans use a map. And if we are in an unfamiliar area we usually drive more cautiously / slower.

 

[EVNow]

Everybody who works in the AV industry understands their value. That's why they use them.

Thats called group think.

The same reason everyone in the auto industry "knew" EVs can't make profit.

 

[JHCCAZ]

Humans use a map. And if we are in an unfamiliar area we usually drive more cautiously / slower.

There's no question that the car needs to use a map for navigation. In fact, it seems to me from watching the videos that some fair portion of problems are caused by incorrect navigation maps. I'm concerned that the dependency on OpenStreetMaps, though it sounds cool to be crowd-sourced and all, is limiting the performance of FSD at the moment (not the only limit of course but an important one). Not only is it doubtful that OpenStreetMaps will ever be reliable enough for the task, it seems very unclear how quickly Tesla will update the map after users make corrections. And how many problems are calls because the corrections are not done correctly?

So let's not let this particular sub-thread devolve into a pitched battle about Maps Good or Maps Bad. I think the debate is about whether the navigation and path planning relies on very high precision mapping of curbs, lanes and myriad other elements of infrastructure. While it does seem tempting to have the HD map data, it's a legitimate question as to how much priority should be placed on the map versus the perceived scene in real time. It's the familiar question: when there is a conflict between the HD map and what is seen, which do you rely on? If you think this through for a while it becomes much less clear that the HD map is very useful at all.

Not long ago, @diplomat33 posted some discussion about MD maps that seemed sensible. I do think Tesla has a map problem as I mentioned above, and I would hope that they would either take advantage of one of these practical high quality (if not High Definition) databases. Or alternatively, launch a much more serious effort to develop Tesla Maps database that is for higher quality than what FSD is attempting to use right now.

 

[powertoold]

HD maps are stupid. Few people think very deeply about what it takes to create them. HD maps need to define everything from practical speed limits to stop lines and where to peak.

Even the process of creating the raw HD maps (without manual labels) isn't straightforward. HD maps are NOT a single pass process. They require multiple passes on multiple days / times. In a city like SF, where you have parked cars occluding road geometry, you may need humans manually extrapolating areas of the map that are constantly occluded by cars or other objects.

 

[powertoold]

This guy is very excited to ride in a Waymo. Amazing technology. What I noticed is how much progress Tesla has made with 10.2's perception. I think 10.2's car perception is almost on par with Waymo. Notice Waymo's car perception is also jittery, similar to 10.2:

 

[stopcrazypp]

Humans use a map. And if we are in an unfamiliar area we usually drive more cautiously / slower.

Definitely humans use maps also, from previously printed maps (or memorizing a set of directions), to GPS navigation, to now navigation that gives lane guidance. Tesla's approach definitely uses maps also, the standard navigation maps is used, Tesla has/had lane maps also (although it's not clear they still use them much, but they did back in 2016), and they also have maps of street signs, signals, and speed limits, although they appear to give only rough locations (and the car still relies on camera to actually confirm what is written and exact position).

Tomtom calls the maps Tesla is using "ADAS map". I previously used the term "MD Map". It's basically a map more advanced than your typical navigation map, and gives accuracy measured in meters. The "HD Map" instead is centimeter accurate and gives info on practically everything on the road, enough that if the car is "synced" to it, and presuming no changes and other objects in the road, the car can essentially drive on the road "blind" just relying on such a map.
What are the differences between the TomTom ADAS and HD Maps? | TomTom Blog

Tesla obviously is not using the latter approach, as you can see in the FSD roadway visualization, that the roadway perception changes in real time and there is a degree of uncertainty. The ones that do a "sync" to a pre-existing detailed map is fairly obvious in that the roadway map is rock solid and doesn't change. As others point out, Tesla is thinking to create a system that can work on general roads, eventually the perception system would need to be able to figure out the road structure just from looking at it, like a human does (even on familiar roadways you rely on your eyes to tell where curbs are, not the map), and Tesla is focused on that part. It could be that they are barking up the wrong tree, and it's not solvable with current tech (so the best is still syncing to a cm accurate map), but I guess we'll find out.