FSD competition is here

[Moderatefan]

Waymo's driverless taxi service details revealed in DMV application

So, Google seems to be beating Tesla to the driverless taxi network.
In Q1 call Musk said they expect to be ready technically by EOY 2019.

Waymo claims to have Level 4 capability now and applied to DMV to operate the first 52 vehicles, hoping to operate 20K in 2 years.

Tesla will need to play a catch up game here.

One interesting thing is that they plan to test in the heavily-mapped area. So, FSD will depend on maps being up-to-date. I assume Tesla will want to use someone else's maps, rather than create their own from scratch. Will Google though want to be liable, if someone else's FSD car jumps off the cliff due to incorrect maps? Interesting how this will get resolved.

 

[voip-ninja]

A lot of people who are more knowledgeable about the technology have insisted that level 4 and up FSD would have to take place initially in urban areas in which every minute detail of the environment was mapped in excruciating detail in order for the car to have to make fewer split second decisions about what to "expect" when navigating the area.

Seems like a perfect marriage for a taxi service that will only operate in a specific urban zone.

Musk continues to be the only one who insists that cameras can do the job that everyone else is relying on LIDAR to do;

What Is Lidar, Why Do Self-Driving Cars Need It, and Can It See Nerf Bullets?

People who paid Tesla $3,000 for full self-driving might be out of luck

The most obvious thing missing from Tesla's cars, from an autonomy perspective, is lidar. The companies that have made the most progress toward fully self-driving cars—including Waymo, Uber, and GM's Cruise—all have lidar on their cars.

Defying the industry consensus, Tesla CEO Elon Musk has repeatedly insisted that lidar is merely a "crutch" and that it's possible to build fully autonomous vehicles using only cameras and radar.

But most industry insiders believe lidar plays an important—and probably essential—role. Cameras offer high range and resolution, but they're not very good at estimating distances and they don't work as well in low-light conditions. Radar provides precise distance and velocity measurements but at very low resolution.

 

[Runt8]

A lot of people who are more knowledgeable about the technology have insisted that level 4 and up FSD would have to take place initially in urban areas in which every minute detail of the environment was mapped in excruciating detail in order for the car to have to make fewer split second decisions about what to "expect" when navigating the area.

Isn't the difference between level 4 and level 5 the fact that level 4 is sandboxed to certain areas?

 

[voip-ninja]

Isn't the difference between level 4 and level 5 the fact that level 4 is sandboxed to certain areas?

Not necessarily from what I understand. It could be something like "road type" vs specific geographic areas.

Example is the car might be able to navigate highways and city streets if they are a certain width or have a certain configuration, but would require the human pilot to take over if turning onto a street that exceeded the system capabilities.

Path to Autonomy: Self-Driving Car Levels 0 to 5 Explained | Feature | Car and Driver

 

[kengchang]

Human ain't got LIDAR

 

[voip-ninja]

Human ain't got LIDAR

Using that comparison, the Tesla Model 3 doesn't have human brain's processing power, which is estimated to be roughly 1 ExaFlop, the current Nvidia platform in the Model 3 is good for about 1% of that raw processing capability.

Right back at ya fella.

 

[brkaus]

Human ain't got LIDAR

And some pretty questionable neural nets...

 

[nvx1977]

As far as AI goes, there's two types here: narrow intelligence and general intelligence. Narrow AI is focused on very specific tasks, like driving, or playing chess. General AI is an open function to be smart at anything you throw at it. Musk believes narrow AI is sufficient for FSD. It's commonplace for narrow AI to surpass humans these days. This is why he thinks LIDAR is a crutch. He's basically betting the ranch on that narrow AI neural net. Very risky and also uncertain when the narrow AI is finally on par with that sliver of our intellect when it comes to driving.

 

[Moderatefan]

Isn't the difference between level 4 and level 5 the fact that level 4 is sandboxed to certain areas?

I'm not positive its about areas, it seems the latest guidance from NHTSA was from 2017
Technology & Innovation

This 2.0: Vision for Safety pdf https://www.nhtsa.gov/sites/nhtsa.dot.gov/files/documents/13069a-ads2.0_090617_v9a_tag.pdf
States Level 4 is

The vehicle is capable
of performing all
driving functions
under certain
conditions. The driver
may have the option
to control the vehicle.

Maybe they can say it doesn't operate in snow. As long as deficiencies are clear.

 

[GatorGuy]

Human ain't got LIDAR

Computers also always pay attention, on everything, 360 degrees and never get tired. They can respond quicker and more exact than any human.

That said, I wish that Tesla was going overboard with redundancies. I would like to see LIDAR, FLIR and multiple forward facing cameras. Rather than seeing how little they can accomplish FSD with, they should be trying to make it as safe as possible and not "safe enough".

 

[kengchang]

Using that comparison, the Tesla Model 3 doesn't have human brain's processing power, which is estimated to be roughly 1 ExaFlop, the current Nvidia platform in the Model 3 is good for about 1% of that raw processing capability.

Right back at ya fella.

Do you use 100% of your brain power to drive? Doubt it

 

[Moderatefan]

Musk continues to be the only one who insists that cameras can do the job that everyone else is relying on LIDAR to do;

I think this bundling of "everyone else" as "advanced" is inappropriate.

The most obvious thing missing from Tesla's cars, from an autonomy perspective, is lidar. The companies that have made the most progress toward fully self-driving cars—including Waymo, Uber, and GM's Cruise—all have lidar on their cars.

It looked like Uber had a pretty bad progress based on articles following the fatal accident. Their system was asking the driver to take control all the time, regardless of Lidar.

 

[Moderatefan]

Rather than seeing how little they can accomplish FSD with, they should be trying to make it as safe as possible and not "safe enough".

Problem is, they had no idea what challenges they may have at the time they installed hardware. This was shot in the dark, sort of.
Once they have a chance to test FSD, it'll be more clear if it's safe enough. Maybe it's fine in 99% of cases and if something breaks, the car can safely stop to satisfy this from guidelines:

Fallback (Minimal Risk Condition)
Entities are encouraged to have a documented process for transitioning to a minimal risk condition when a problem is encountered or the ADS cannot operate safely. ADSs operating on the road should be capable of detecting that the ADS has malfunctioned, is operating in a degraded state, or is operating outside of the ODD. Furthermore, ADSs should be able to notify the human driver of such events in a way that enables the driver to regain proper control of the vehicle or allows the ADS to return to a minimal risk condition independently.

 

[roblab]

Using that comparison, the Tesla Model 3 doesn't have human brain's processing power, which is estimated to be roughly 1 ExaFlop, the current Nvidia platform in the Model 3 is good for about 1% of that raw processing capability.

Right back at ya fella.

Mebbe so, but you KNOW that a goodly portion of that exaflop is not worried about driving, but about what's on the radio, what the significant other said last night, what's for lunch, who's driving that cool red car up ahead, whether there's a cop around the next bend, and whether or not I ought to floor it past the kid in the Corvette. Nvidia only thinks about boring stuff, like driving, which is probably only about 1% of what I'm thinking of.

 

[Moderatefan]

Mebbe so, but you KNOW that a goodly portion of that exaflop is not worried about driving, but about what's on the radio, what the significant other said last night, what's for lunch, who's driving that cool red car up ahead, whether there's a cop around the next bend, and whether or not I ought to floor it past the kid in the Corvette. Nvidia only thinks about boring stuff, like driving, which is probably only about 1% of what I'm thinking of.

Plus humans are bad at multi-tasking, so all this thread management must be eating up at least 50% of the available capacity.

 

[Tezlanian]

Plus humans are bad at multi-tasking, so all this thread management must be eating up at least 50% of the available capacity.

Humans multitask constantly and parts of the brain work almost autonomously just reporting to results to higher executive function. It's that top level that doesn't multitask well. Those ares are a much small percentage of overall brain.

And yes, I knew you were joking

 

[voip-ninja]

Do you use 100% of your brain power to drive? Doubt it

Do you use more than 1%? I know I do. Much more than 1%.

I used to ride sport bikes in the canyons. Uses similar amount of brain power as flying performance aircraft, which is a LOT.

Nobody who rode motorcycles for years and a hundred thousand miles surrounded by distracted idiots in cages trying to kill them drives casually around using 1% of their brain.

 

[kengchang]

Do you use more than 1%? I know I do. Much more than 1%.

I used to ride sport bikes in the canyons. Uses similar amount of brain power as flying performance aircraft, which is a LOT.

Nobody who rode motorcycles for years and a hundred thousand miles surrounded by distracted idiots in cages trying to kill them drives casually around using 1% of their brain.

So do you need 1 ExaFlop or not? If not quoting that number is pointless

 

[Stolz25]

Do you use 100% of your brain power to drive? Doubt it

Of course not, I need at least 75% to type out responses on forums, mayebejsjjduisjsi

 

[insaneoctane]

Do you use 100% of your brain power to drive? Doubt it

No way! You know how many times I've gotten to my destination and couldn't remember anything about the drive because I was driving as a "background" task?