Waymo will beat Tesla FSD because of LIDAR

[Deleted member 17810]

Dojo uses machine-learning to create an AI model, which then runs on the FSD computer. Neither is "generating" new content.

You might make the case that it's "generating" the on-screen visualization, but that's really just a rendering of the data, and is not necessary for FSD to function.

Dojo is using camera data to analyze physical locations and rebuilding 3d models to put back into the self driving model. This is in their white paper. This is generative AI and machine learning.


It's why some locations are better driving than others. It has more data. The more a place is driving the "better" is behaves. It's also why construction zones and new areas cause it to just revert to old patterns.

Sponsored

 

[Crissa]

What do you think dojo is doing with all the visual image to spit back out to the cars to traverse through?

Using generative would produce insane results. We want it to drive in the real world, not an imaginary one.

AI that knits together video data into 3D worlds isn't generative.

-Crissa

 

[CyberTW]

*Disclaimer: I work for Google, but not Waymo
*Overall opinion: I use the Cybertruck's FSD all the time, even to drive 2-minutes down the street; it is awesome because it's here NOW

After seeing many FSD threads this past week and enjoying the Cybertruck's FSD myself I wanted to share a brief crash course on autonomous driving

References:

• MIT's course on Artificial Intelligence: Implications for Business Strategy
• Lecture, "Understanding Robotics"
• Lecture transcripts attached for convenience

Robotics 101: AI in the Physical World, with Sensors and Actuators
At a high level robots are just automated devices that perform physical tasks in the real world using 1. sensors (inputs) and 2. actuators (outputs)

Tesla is Level 3 Autonomy, Waymo is Level 4

All Self Driving Cars "Map" Between Feature Space (virtual) and Physical Space (real)

• All of the sensors (cameras, LIDAR, radar, microphones, etc.) are used to create a map of our real world in a virtual world called "feature space"
• The car "thinks" and "acts" entirely in this feature space
• It needs the sensors to keep populating the feature space with objects, to try to mimic our physical world as much as possible
• This why the Cybertruck "drives over curbs" while Waymo "sees the curbs", Waymo's feature space is better populated, closer to our real world

Tesla FSD is Parallel Autonomy (human-in-the-loop), Waymo is Series Autonomy (no human, mostly)
In robotics there are two types of autonomy, series and parallel:

• Series autonomy (Waymo): either human in control or vehicle in control, not both
• Parallel autonomy (Tesla FSD): aka "Supervised" FSD

Tesla's Big Failure: No LIDAR
In one sentence, robots need "laser imaging, detection and ranging" (LIDAR) to accurately perceive depth, occlusion, etc.

Longer answer is that LIDAR facilitates building a 3D point map of the world. Initially cameras were meant only for object detection. Tesla admirably expanded camera-based computer vision into FSD, but the lack of LIDAR is still problematic

Tesla's Reason for Camera-Only Computer Vision FSD: Cost
LIDAR (and the now removed radar from early model Teslas) is expensive and it's not just the hardware. The Cybertruck FSD update boasts that its neural network / machine learning model replaces over "300K lines of explicit C++ code". That's a lot of software engineering hours and millions of dollars saved

My (unverified) opinion is that there are nowhere near as many LIDAR 3D map examples as there are computer vision image and video examples. This means training a LIDAR machine learning model is not (yet) feasible

Waymo cleverly overcame this limitation by transforming the LIDAR 3D point map into a more manageable format. I have not had the time to read all the research, but Waymo solved the LIDAR point map problem, where a mess of 3D dots inhibited using machine learning fully

All of Waymo's research came at an extreme cost and they had Google / Alphabet money. Tesla just couldn't afford to run at a loss and had no other massively profitable business units to draw from

When waymo can get out of its geofence, I will take this seriously… I have had my Tesla drive me 240 without touching a steering wheel/pedal or brake.. through freeway, highway, city streets and non-painted line streets… tell me waymo can do that right now and I will believe you

 

[HaulingAss]

I don't think that fsd is bad science or that it's bad, that's something that you and alot of people are attributing to my comments.

Well, you said Waymo is better. That makes FSD bad!

Obviously.

 

[Crissa]

When waymo can get out of its geofence, I will take this seriously… I have had my Tesla drive me 240 without touching a steering wheel/pedal or brake.. through freeway, highway, city streets and non-painted line streets… tell me waymo can do that right now and I will believe you

Waymo won't drive out of its geofence because of liability, not ability.

Waymo is taking the liability when its AI driver does something wrong. Tesla isn't.

-Crissa

 

[CyberTW]

Waymo won't drive out of its geofence because of liability, not ability.

Waymo is taking the liability when its AI driver does something wrong. Tesla isn't.

-Crissa

Ok.. so when/if Tesla does… Waymo does me no good.. if they can’t scale and take on liability outside of a minuscule area maybe it isn’t as great of a system as they think

 

[CyberTW]

Waymo won't drive out of its geofence because of liability, not ability.

Waymo is taking the liability when its AI driver does something wrong. Tesla isn't.

-Crissa

It’s like saying you have the best phone on the market (and that may be right), but you say you can only use it in a 1x1mile radius.. cool, good for you for developing the best system that doesn’t work barely anywhere

 

[CyberGus]

Dojo is using camera data to analyze physical locations

Yes!

This is generative AI

NO! "Generative AI" is a specific term, to mean "creating new art".

Creating a picture from the prompt "make a picture of kangaroos dressed as pirates playing badminton" is "Generative AI". Tesla, OTOH, is using a large data set (vehicle video and telemetry) to create an AI model of Neural Networks. The model is algorithms and probabilities, there's no actual imagery within it.

 

[cyberos]

It’s like saying you have the best phone on the market (and that may be right), but you say you can only use it in a 1x1mile radius.. cool, good for you for developing the best system that doesn’t work barely anywhere

Remember when iPhones couldn't do MMS and then couldn't do animated gifs, but every other phone could?

 

[REM]

That's fine, but until teslas vision can repeatedly demonstrate the vision only camera can detect distances, then it's not safe.

Cameras absolutely can calculate distance accurately; who has convinced you that it doesn't? lol

 

[REM]

LIDAR typically uses infrared light, and is not visible.

Still, it begs the question of interference between multiple LIDAR units, or if some joker with a laser pointer could get you stuck in traffic lol

Yep, I'm aware I have taken a few trips in a Waymo and never saw any laser glints (can see the reflectors spinning though, which is pretty cool).

I was speaking from the perspective of the sensors seeing scatter from other lidars, so it looks like you caught on to what I was getting at. You could probably have the lidars use spread spectrum techniques to help prevent waveform interreference between other cars, but I'm sure those lidar units already restricted to a very narrow band in the first place. So probably not enough space between frequencies to hop over to another channel.

 

[cyberos]

Yes!



NO! "Generative AI" is a specific term, to mean "creating new art".

Creating a picture from the prompt "make a picture of kangaroos dressed as pirates playing badminton" is "Generative AI". Tesla, OTOH, is using a large data set (vehicle video and telemetry) to create an AI model of Neural Networks. The model is algorithms and probabilities, there's no actual imagery within it.

They are both GenAI ?

• Imagen 3 prompt: Make a 4k photo of a female karate master in a dojo, using a video camera like a weapon or tool, to build a wireframe image of a Cybertruck
  
  
• Dall-E 3 prompt: Make a picture of kangaroos dressed as pirates playing badminton

I'm off to the "We, Robot watch party!

If they don't have LIDAR in the Robotaxi then you can assume I have fled the Internet in shame

 

[Crissa]

Remember when iPhones couldn't do MMS and then couldn't do animated gifs, but every other phone could?

No, I do not remember that. Because none of my phones ever could do that before an iPhone could. (I used my computer for that stuff.)

-Crissa

 

[REM]

Dojo is using camera data to analyze physical locations and rebuilding 3d models to put back into the self driving model. This is in their white paper. This is generative AI and machine learning.

It's why some locations are better driving than others. It has more data. The more a place is driving the "better" is behaves. It's also why construction zones and new areas cause it to just revert to old patterns.

Precisely! Dojo is probably one of the coolest and most impressive simulation compute clusters on the planet. I would LOVE to see what their real-time 3D mapping looks like. I bet you can explore it like you can google maps/street view.

 

[dalton108]

They are both GenAI ?

• Imagen 3 prompt: Make a 4k photo of a female karate master in a dojo, using a video camera like a weapon or tool, to build a wireframe image of a Cybertruck
  
  
• Dall-E 3 prompt: Make a picture of kangaroos dressed as pirates playing badminton

I'm off to the "We, Robot watch party!

If they don't have LIDAR in the Robotaxi then you can assume I have fled the Internet in shame

au revoir!

Sponsored