Autonomous Car Progress

[Bitdepth]

Oh you don't have to worry about that we know you're anti Tesla.
And you've proven time and again that the one naive and arrogant is you, not Musk.

You are out of your mind. It was only a few years ago that @diplomat33 was arguing your position and then he just like everyone who is interested in Tesla and autonomous cars came to the same realization that Tesla is not solving FSD any time soon. The goal post has been moving since 2016.

We went from this 2016

To this 2019

And promises of pending regulation. If you look at both videos you would conclude that tesla should have at least a level 3 self driving car by now wouldn't you? Now ask yourself why hasn't any of those video manifested in actual customer vehicle even though tesla is known to push beta software out to customers. News flash, having a realistic view on things does not make you anti tesla especially people who have spent money buying tesla products and evangelize the benefit of owning tesla vehicles.

 

[powertoold]

Nothing in Elon's past matters imo.

Fact is no one is closer to level 5 than Tesla.

You can't simulate real world data, especially for edge cases. Anyone working with NNs knows that.

 

[Nuclear Fusion]

I still believe in Musk’s tweet of 24th January 2017 that it will be “3 months maybe, 6 months definitely“

 

[Kant.Ing]

I dug out some more...

10/10/14 Bloomberg interview: "And so I think from the point at which true autonomous driving is possible which I now think is probably the five or six year timeframe. I thought was more like 10 years. But based on the rate of improvement that I'm seeing and the progress we're making I think we'll probably do it in five or six years and then it's as follows looking to take two or three years after that before the regulatory regulators approve it."
"I think we’ll be able to achieve true autonomous driving, where you could literally get in the car, go to sleep and wake up at your destination,”

4/28/2017 TED Talk: Tesla Model 3 is coming in July, Musk says, and it’ll have a special feature: autopilot. Using only passive optical cameras and GPS, no LIDAR, the Model 3 will be capable of autonomous driving. “Once you solve cameras for vision, autonomy is solved; if you don’t solve vision, it’s not solved … You can absolutely be superhuman with just cameras.” Musk says that Tesla is on track for completing a fully autonomous, cross-country LA to New York trip by the end of 2017. “November or December of this year, we should be able to go from a parking lot in California to a parking lot in New York, no controls touched at any point during the entire journey,” Musk says.

7/8/2020 World AI Conference: “I remain confident that we will have the basic functionality for level five autonomy complete this year,” Musk said. “I think there are no fundamental challenges remaining for level five autonomy. There are many small problems, and then there’s the challenge of solving all those small problems and then putting the whole system together, and just keep addressing the long tail of problems.”

It looks like his 10 years prediction is the best shot among all so far. So 2024, when TSLA hit 10,000?

 

[diplomat33]

Nothing in Elon's past matters imo.

WOW. So you would be ok even if Elon BS'ed and lied since it would be in the past?

Fact is no one is closer to level 5 than Tesla.

That is not a fact. That is your opinion. Many companies like Waymo already have L4 and Tesla is L2. So, they are much closer to L5 than Tesla is. Having "autonomous driving that works somewhere" (L4) is closer to "autonomous driving that works everywhere" (L5) than "driver assist that works in some conditions" (L2).

 

[DanCar]

... closer to level 5 than Tesla. You can't simulate real world data, especially for edge cases. Anyone working with NNs knows that.

That is a narrow view of what it takes to achieve level 5. A wider view is that you will need general A.I. . Nobody has that, and no one will in the next 5+ years. Levels 3 and 4 are easier to achieve without common sense. I do believe Tesla will achieve level 3 before Waymo has wide deployment of their service a few years from now.

 

[powertoold]

That is a narrow view of what it takes to achieve level 5. A wider view is that you will need general A.I.

I think that overly complicates driving. You don't need general AI for level 5. Almost all driving involves repetition. Sure, there might be a few situations where general AI would help, but for those 0.0001% of trips, Tesla can develop a contingency feature.

The most complicated "reasoning" you do while driving is "can I fit through there", "should I wait behind or go around", "is the guy waving me to pass or stop?" Otherwise, most of non-routine driving just involves avoiding accidents and anticipating the pattern of other moving objects.

 

[DanCar]

but for those 0.0001% of trips

If you are not handling all cases then you are talking about level 4, not 5 per definition. And I wouldn't be so generous with those zeros. Adversarial attacks will come, from teenagers just to show how dumb the cars are. Heavy rain, snow, etc... isn't that uncommon.

 

[diplomat33]

The most complicated "reasoning" you do while driving is "can I fit through there", "should I wait behind or go around", "is the guy waving me to pass or stop?"

Keep in mind that driving can often require solving multiple problems at the same time. So you might need to solve "is the guy waiving me to pass?" and "what is that car coming towards me going to do?" and "is that pedestrian going to cross?" all at the same time. So the issue is not how difficult one single problem is but solving multiple problems at the same and knowing which problems to solve first.

The thing about driving is that the level of complexity actually changes in both space and time, meaning there are locations and times where driving is less complex and locations and times where driving is more complex. For example, cruising down a well marked empty interstate highway in broad daylight requires very few driving tasks (just stay in the lane and maintain a safe speed). But if you detect a car in front of you that is driving slower than you or you detect a construction zone ahead, the number of driving tasks will go up. If you are stuck in traffic and not moving, the number of driving tasks needed in that instant, temporarily drops to zero but increases again when traffic starts moving. And in an urban setting with pedestrians, cyclists, dozens of other vehicles of all shapes and sizes, intersections with traffic lights, the number of driving tasks can become quite high.

On a related note, that's probably why companies like Waymo and Cruise are focusing exclusively on dense urban driving. It's the most difficult driving to solve. Once you solve dense urban driving, the other driving scenarios will be easier to solve because they are less complex. Also, dense urban areas will have the most demand for an autonomous taxi service so it makes sense for ride-hailing services to focus on that first.

 

[MarsNext]

Love my Model 3 but it is a little difficult to explain to non-Tesla owners why you paid $7,000 for FSD

 

[2101Guy]

Nothing in Elon's past matters imo.

.

I mean, some believe that a man's word is a crucial component of one's character and overall reputation..

 

[DanCar]

I mean, some believe that a man's word is a crucial component of one's character and overall reputation..

Ok a man does or doesn't have character or reputation, but does the system work?

 

[powertoold]

I mean, some believe that a man's word is a crucial component of one's character and overall reputation..

Yup, my statement was with regards to his fsd predictions. I think if Tesla actually finishes feature-complete FSD this year (in releaseable form or not), Elon should be forgiven for his past hyperbolic FSD statements.

I, like many currently, used to undervalue the significance of feature-complete (basic functionality) FSD. It's a bit confusing since we have all these other autonomous car companies that seem to have achieved "feature-complete" already. That why it seems like Tesla is so far behind and so why should we applaud Tesla for just catching up?

Lately, I've been getting a better understanding of what Tesla is trying to achieve, and to me, the ambition is mind-blowing. Elon's predictions have been very good considering the ambition, imo. It just takes some knowledge of NN research and the underlying difference in Tesla's approach.

 

[Nuclear Fusion]

I mean, some believe that a man's word is a crucial component of one's character and overall reputation..

True, except for Elon Musk. See anything Musky says that was, well, a crock of *sugar*, can be forgiven, ignored, etc. Like that there will be 1 million robotaxis on the road at the end of this year. Or that we will be able to sleep in our cars on the drive to work this year. Or that definitive statements are subsequently qualified with the words ‘feature complete’, known in politics as ‘double speak’ as it fools the less critical thinking.
See, so many obsequious acolytes have bound their self image to Musk, that resolving cognitive dissonance is already plugged into one’s daily planner. But, but, but on current trends cases/instances of such will soon be close to zero.
Additionally, I have it on good information that universities are soon closing worldwide & the peer review process is being ceased and we will simply await the latest Musk tweet for our education, since a Wolowitz, oops, I mean engineer is the fount of all knowledge

 

[diplomat33]

I think this is pretty neat. It shows Nvidia Drive's localization using camera vision and HD maps. On the left, we see the camera vision detecting lane lines and road signs and then it matches it to the HD map, seen on the right, to localize the car with cm level accuracy. And we see it works in various weather conditions.

 

[diplomat33]

I think this video is very informative about HD maps. I am sharing it for educational purposes.

I certainly learned something new. For example, I did not know that HD maps actually annotate the "waiting areas" where it is safe to stop when in the middle of an unprotected left turn. Also, by annotating the safe paths each lane can take, it helps the car predict what other cars can do at an intersection which is very useful in making the driving safer.

Again, I am not saying that HD maps can replace camera vision but they certainly can provide a lot of useful information that can help make autonomous driving safer.

The video also notes that in 2019, NavInfo completed HD maps for all the highways in China. If you can do HD map all the highways in a country the size of China, then I think HD maps can work for L5 autonomy.

 

[mikes_fsd]

Yes, we should use concrete definitions. Luckily, the SAE already provides us with some.

There is only one problem with SAE classes for FSD.
They do not specify / quantify who the feature set is available to.
A university or a company, like Apple (or any other cash rich institution), with $#!t ton of money can pay for R&D for years and produce a handful of L4 or L5 cars... BUT those cars never get in to the hands (or be used by) the general public.

To me, this is a very important metric, "who is able to use the system".

for example:
A L3 car that is available to anyone for a few $k on their vehicle is 1000s of times more useful than a L5 car that is only used for promo videos or in a university research project.

 

[diplomat33]

There is only one problem with SAE classes for FSD.
They do not specify / quantify who the feature set is available to.
A university or a company, like Apple (or any other cash rich institution), with $#!t ton of money can pay for R&D for years and produce a handful of L4 or L5 cars... BUT those cars never get in to the hands (or be used by) the general public.

To me, this is a very important metric, "who is able to use the system".

for example:
A L3 car that is available to anyone for a few $k on their vehicle is 1000s of times more useful than a L5 car that is only used for promo videos or in a university research project.

Sure those are important metrics for the consumer to consider when buying a car but they have nothing to do with whether the tech is autonomous or not. From the point of view of the SAE levels, they are off topic. The SAE levels are only interested in defining what autonomous driving is. So the SAE levels focus on describing the characteristics of the autonomous tech. For example, can the autonomous car handle a failure or does it need to ask the driver to intervene? Can the autonomous vehicle handle stopping for a pedestrian? Can the autonomous vehicle operate on all roads or only some roads? Etc... These are critical questions that define the nature of the autonomous tech. That's what the SAE is interested in.

So in the examples you give, the SAE is not interested in assigning a value of which system is "better" in terms of usefulness to the consumer. It is simply interested in defining what the tech can do, what the differences are between a L3 car and a L5 car. Remember that the SAE stands for Society of Automotive Engineers. So it is a society focused on the engineering questions of autonomous driving.

Questions like cost or availability are separate questions for the consumer to consider.

 

[john5520]

From what I understand, Waymo's goal is to make their tech available to as many manufacturers as possible. Think of them like Microsoft Windows. Once the hardware can accommodate it, it'll work.
So if they get their way, they may end up grabbing the lion's share of the self driving market.

Tesla is more like Apple. They make the hardware and software and is more proprietary. And they've got a large faithful customer base and great reputation. They're not going anywhere soon.
However, they're just one player in a growing market. Right now, there is a still space to accommodate others and they don't need to be number one.

As for the SAE level definitions, they should stay as specific as possible. Adding any other criteria would unnecessarily complicate things. Whether it's a couple cars in a city, or all over, shouldn't matter.

 

[BlueBlazes]

WOW. So you would be ok even if Elon BS'ed and lied since it would be in the past?



That is not a fact. That is your opinion. Many companies like Waymo already have L4 and Tesla is L2. So, they are much closer to L5 than Tesla is. Having "autonomous driving that works somewhere" (L4) is closer to "autonomous driving that works everywhere" (L5) than "driver assist that works in some conditions" (L2).

Waymo's L4 is limited to a few geographic areas where it has mapped (and continuously re-maps to catch changes) every street, light, stop sign, traffic sign, etc., down to the millimeter. As I understand it, their LIDAR looks at the surroundings (pre-mapped, remember) and determines the exact location of the car. They use this to essentially drive like they're on rails. They don't need to read lane markers because they're always in the right place based on the maps. But even Waymo recently said they will still need safety drivers for at least a few more years. So that doesn't sound like a true L4 to me. If this isn't correct, I would appreciate it if someone would educate me.