Waymo brings in $2.25 billion from outside investors, Alphabet

[mikes_fsd]

Tesla is not using the same hardware at all. For example, Tesla is not using lidar. Tesla's approach and Waymo's approach are not compatible.

Tesla eliminated a component (that Tesla believes to be unnecessary) but the goal is the same...

 

[diplomat33]

It is OK to disagree @diplomat33
my question still stands:
You bought FSD, what did you think you were getting?

clearly you are not disagreeing with me, but with Elon, who is in charge of the direction of Tesla and the whole Autopilot direction.

I answered your question earlier: I expect to get more advanced driver assist features. Eventually, I think Tesla will add enough features and capabilities that the car will be able to drive itself but with driver supervision.

Yes, I am disagreeing with Elon's approach to FSD. Lidar is needed for safe FSD.

Tesla learns, in fact every Tesla car on the road helps Tesla Inc learn.

Because Waymo's approach is not scalable (they are on the 5th generation of their hardware and every time they claim it is more scalable)
It is also way too expensive still.

But to quote Elon "Lidar is a fool’s errand... Anyone relying on lidar is doomed. Doomed! [They are] expensive sensors that are unnecessary. It’s like having a whole bunch of expensive appendices. Like, one appendix is bad, well now you have a whole bunch of them, it’s ridiculous, you’ll see."

It is fine if you want to believe in Tesla's FSD but please don't write false things about Waymo.

1) Waymo's approach is very scalable.
2) Lidar is not expensive anymore. This Lidar Is So Cheap It Could Make Self-Driving a Reality
3) LIDAR is not a crutch. It is a critical component to do FSD safely.

 

[mikes_fsd]

It is fine if you want to believe in Tesla's FSD but please don't write false things about Waymo.

1) Waymo's approach is very scalable.
2) LIDAR is not a crutch. It is a critical component to do FSD safely.

I could argue that you are writing false things about Tesla FSD.
I have in my corner Elon, Karpathy, Sebastian Thrun and even Anthony Levandowski (yeah the guy who was accused of stealing lidar junk when he left for Uber) -- there are others.

Sebastian Thrun, a self-driving pioneer who led Google’s self-driving car project that eventually spun off as Waymo, led the Stanford Racing Team to a win in the DARPA Grand Challenge with a lidar-equipped car nearly 15 years ago. But a lot has changed since then.

“My opinion has shifted a bit” about whether self-driving vehicles need lidar, said Thrun, founder and president of Udacity, an online education company. He cited the advent of deep learning, which helps cars’ computers make decisions based on information collected by their sensing systems, regardless of which sensors they use.

These are not schmucks that know nothing about Autonomous Driving or what it takes.... so please stop with your "false things about FSD".

 

[Tam]

Why cannot Tesla learn and build on Waymo's technology?

2 different philosophies.

1) Geofencing & Preparation for the area:

Waymo believes in the geofencing system because it works better for an automated system if everything is predictable, every single possible scenario is listed and solved. If it can do it with the highly prepared 50 squared miles in Chandler, AZ, it should be able to expand the same method to other areas too. It just takes lots of time to expand the system to the whole USA.

That method works best in driverless railway in Disney Land where the rail is raised up high to make sure there's no obstacle and the route is unchangeable. It also works well in an elevator where all the scenarios are predicted (geofenced to a specific, unchanging number of floors).

2) General, generic solution:

NTSB wants Tesla to geofence Autopilot so it can't be used in the city as dictated by its owner's manual but Tesla aims at a general solution rather than geofencing.

3) Cameras are eyes:

Tesla believes that because human can drive with 2 eyes only, it should also with 8 cameras!

The problem is: Human does have a brain between 2 eyes so Tesla must develop a similar "brain" with its software.

So the issue is only a matter of time that someone (or some machine) is smart enough to write codes for those 8 cameras.

4) No wait for camera software:

Waymo believes while waiting for advanced software for cameras, it can use a training crutch called LIDAR that it can write practical codes for now.

5) Consumer choices:

Those 2 above thoughts are still imperfect but I am willing to choose Tesla because I can currently use it almost anywhere.

 

[diplomat33]

I could argue that you are writing false things about Tesla FSD.
I have in my corner Elon, Karpathy, Sebastian Thrun and even Anthony Levandowski (yeah the guy who was accused of stealing lidar junk when he left for Uber) -- there are others.

These are not schmucks that know nothing about Autonomous Driving or what it takes.... so please stop with your "false things about FSD".

I have not said anything false about Tesla's FSD. I think I understand their method quite well: develop the software that takes what the 8 cameras see and use that information to drive the car. To do that, they are developing neural networks to recognize objects, measure distance, trace paths etc...

 

[diplomat33]

Waymo believes in the geofencing system because it works better for an automated system if everything is predictable, every single possible scenario is listed and solved. If it can do it with the highly prepared 50 squared miles in Chandler, AZ, it should be able to expand the same method to other areas too. It just takes lots of time to expand the system to the whole USA.

Once Waymo's autonomous driving is good enough, then they will have a generalized solution and they won't need geofencing anymore.

 

[mikes_fsd]

You are saying that Tesla needs Lidar:

Lidar is needed for safe FSD.

That is a FALSE statement.

Also, on solid state lidar, that won't be in cars until 2023 according to that article, we will see what happens.

Once Waymo's autonomous driving is good enough, then they will have a generalized solution and they won't need geofencing anymore.

Once Tesla neural nets are good enough, the FSD package will cost ~5x more then what it costs today!

With Waymo's track record it will be another 10 years before east coast gets to see those cars regularly.

 

[diplomat33]

Why cannot Tesla learn and build on Waymo's technology? What is the fundamental difference?

The two approaches are very different.

Tesla's approach: develop the software to take what the cameras see and use the information to drive the car. To do this, Tesla is developing neural networks to recognize objects, lane lines, etc...

The main problem with this approach: it requires super good computer vision that does not exist yet. And if your cameras get blocked, obstructed or fail, your car can't drive. And even with perfect vision, that is just the first step. After that, you still need to write the rules for how the car will drive.

Waymo approach: Waymo uses a combination of cameras, radar, hd maps and lidar to guarantee that it has an accurate view of the world and objects around the car. It also gives the car a lot of redundancy. If one sensor fails, the car can still drive. Then, with this super accurate 3D view of objects around it, Waymo develops the software to tell the car how to navigate the world and how to react to other objects.

 

[diplomat33]

You are saying that Tesla needs Lidar:

That is a FALSE statement.

No, it is not a false statement. Every leader in FSD says that LIDAR is required because camera vision alone is not yet good enough. Plus, even if it was good enough, you still need redundancy in case the camera fails. You've seen what happens when a camera on our Tesla is obstructed and we get the message that "autopilot is limited". How is Tesla going to do FSD when a camera is obstructed?

 

[mikes_fsd]

I will let Karpathy's words just sit here...

Andrej Karparthy, Senior Director of AI, took the stage and explained that the world is built for visual recognition. Lidar systems, he said, have a hard time deciphering between a plastic bag and a rubber tire. Large scale neural network training and visual recognition are necessary for Level 4 and Level 5 autonomy, he said.

“In that sense, lidar is really a shortcut,” Karparthy said. “It sidesteps the fundamental problems, the important problem of visual recognition, that is necessary for autonomy. It gives a false sense of progress, and is ultimately a crutch. It does give, like, really fast demos!”

After that, you still need to write the rules for how the car will drive.

Ah, yes, Waymo does not have to write any rules, the Lidar just makes the car drive... lol.
Everyone working on the autonomy problem has to write rules.

The real difference between Waymo and FSD is that Waymo insist on using Lidar.

 

[mikes_fsd]

How is Tesla going to do FSD when a camera is obstructed?

An autonomous car can have brakedowns.
If enough sensors go out, the car should come to a safe stop (slow down with hazard lights on and come to a stop)

But if it is not sensor failure then sensor redundancy should be enough.
Tesla is not yet using all camera sensors as a fallback (they are all on and getting data, but not yet able to be used as fallback)

 

[diplomat33]

I will let Karpathy's words just sit here...

"Andrej Karparthy, Senior Director of AI, took the stage and explained that the world is built for visual recognition. Lidar systems, he said, have a hard time deciphering between a plastic bag and a rubber tire. Large scale neural network training and visual recognition are necessary for Level 4 and Level 5 autonomy, he said.

“In that sense, lidar is really a shortcut,” Karparthy said. “It sidesteps the fundamental problems, the important problem of visual recognition, that is necessary for autonomy. It gives a false sense of progress, and is ultimately a crutch. It does give, like, really fast demos!"

Karpathy is saying that because he and the rest of Tesla believe that they can get a super good computer vision that will be able to create that accurate 3D view of the world without lidar.

Ah, yes, Waymo does not have to write any rules, the Lidar just makes the car drive... lol.
Everyone working on the autonomy problem has to write rules.

I never said that. No, lidar does not just make the car drive. Lidar just helps you create that accurate 3D map to solve the perception part. Waymo still spent years writing the rules.

The real difference between Waymo and FSD is that Waymo insist on using Lidar.

Perhaps, but thanks to the "crutch" of lidar, Waymo can do FSD already and Tesla can't do FSD yet until they finish computer vision.

 

[mikes_fsd]

Perhaps, but thanks to the "crutch" of lidar, Waymo can do FSD already

Waymo's FSD is like a ride at Disney, defined route in a defined place.... but you can still get stuck in traffic. That is NOT an FSD that I am looking for.

 

[mikes_fsd]

Karpathy is saying that because he and the rest of Tesla believe that they can get a super good computer vision

Waymo keeps pushing Lidar because they do not want to admit that they've wasted 10 years on Lidar because they failed to understand the fundamental problem of vision (in the sense of understanding what you are seeing) for autonomous driving.

 

[VT_EE]

Will someone point to a LIDAR sensor that works in rain and snow as well as a passive sensor such as a camera? I’m one who doesn’t buy into the idea that LIDAR is necessary for FSD.

 

[diplomat33]

Tesla is doing a big rewrite of their FSD. I hope it works and gives us good features. I am just skeptical of Tesla's approach because it requires that the computer vision be flawless. If the computer vision makes any mistake at all, it can be deadly, as we've seen with the Tesla crashes and you can't remove driver supervision until it is perfect and the driver policy is finished. And there is no redundancy if the computer vision makes a mistake.

 

[diplomat33]

Waymo keeps pushing Lidar because they do not want to admit that they've wasted 10 years on Lidar because they failed to understand the fundamental problem of vision (in the sense of understanding what you are seeing) for autonomous driving.

That's the kind of total BS that I wish you would stop posting.

 

[diplomat33]

Will someone point to a LIDAR sensor that works in rain and snow as well as a passive sensor such as a camera? I’m one who doesn’t buy into the idea that LIDAR is necessary for FSD.

LIDAR will work in light rain and light snow. Cameras will fail in heavy rain and snow too. Just think about your vision. How well do you see in heavy rain, dense fog or snow. Answer: not well.

Cameras don't work well at night whereas lidar works great at night. Radar is excellent in rain and snow.

That is why you need cameras, radar and lidar together. When you combine all three sensors together, each sensor will compensate when another sensor can't see well, and you will get a good 3D map of the world in all weather conditions

 

[mikes_fsd]

That's the kind of total BS that I wish you would stop posting.

The only reason I even post here, is because of you "total BS" about need for lidar.

Will someone point to a LIDAR sensor that works in rain and snow as well as a passive sensor such as a camera? I’m one who doesn’t buy into the idea that LIDAR is necessary for FSD.

There is no hardware way to do it.
There is a way to get better at it, but basically you are writing "vision for Lidar"

Here is an article that describes an approach...:

“If you record not just the first thing your laser hits, but subsequent things, including the last thing, you can reconstruct a whole ground plane behind what you’re seeing, and you can infer that a snowflake is a snowflake,” he told Quartz.

Additionally, the algorithm checks for the persistence of a particular obstacle. A laser beam is unlikely to hit a raindrop twice, for example, allowing the algorithm to rule it out as an obstacle, McBride told Quartz.

Driverless cars have a new way to navigate in rain or snow

But now, you are writing software, to do vision specific to Lidar, so instead of focusing your energy on the root problem of vision, you are fragmenting your team into solving the Lidar vision problems and camera vision.

 

[mikes_fsd]

LIDAR will work in light rain and light snow.

That is yet another false statement.

Without writing specific code to handle this, you will get false positives on obstacle detection.
You end up literally writing vision for lidar...