Autonomous Car Progress

[EVNow]

I'm all for another Urban Challenge involving every ADS company except in the real world traffic. Most importantly no one in the driver's seat, maybe someone in the passenger seat with a single RED disengage button that would stop the car in its track.

That is not practical or useful at this point.

Waymo vs FSD beta is both possible and interesting. I’m sure we can get some people to try it out and post on YouTube.

 

[Bitdepth]

That is not practical or useful at this point.

Waymo vs FSD beta is both possible and interesting. I’m sure we can get some people to try it out and post on YouTube.

It would be more difficult to organize but it's nothing that hasn't been done before. There was one in China a few months ago. It would be very useful to see how the state of the art in the industry deals with real-world traffic and problems in the same geographical area.

 

[powertoold]

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.

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

 

[Bitdepth]

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

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

 

[powertoold]

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

Sorry, I just don't find your commentary convincing. It's an issue until it's fixed or reduced to the point of being irrelevant. The day will come.

If you can point out a problem that Tesla may NEVER fix, then I'm all ears.

 

[Bitdepth]

Sorry, I just don't find your commentary convincing. It's an issue until it's fixed or reduced to the point of being irrelevant. The day will come.

I'm not trying to convince you of anything. I'm stating a fact. Not running into objects is the number one priority for an autonomous vehicle and it is not a "minor" issue as you put it.

 

[powertoold]

I'm not trying to convince you of anything. I'm stating a fact. Not running into objects is the number one priority for an autonomous vehicle and it is not a "minor" issue as you put it.

Here's the typical strategy for your type of logic:

1) New FSD beta version releases, improves some static object response, doesn't improve others
2) Ignore things that were improved, point out things that weren't
3) Rinse and repeat until the logic no longer works and then be quiet

 

[Terminator857]

I'm not trying to convince you of anything. I'm stating a fact. Not running into objects is the number one priority for an autonomous vehicle and it is not a "minor" issue as you put it.

When powertoold states it is a minor issue, he means it is relatively easy to fix. I tend to agree. Yes, it is a major issue in terms of safety but that is not the point he is trying to make. He is saying the issues he sees are easy to fix, thus minor.

Tesla has already declared that HW3 will be insufficient for robotaxi. My bet is that in 4 or so years, HW4 will also be declared insufficient for full self driving. That doesn't mean that HW4 won't accomplish a ton.

 

[Bitdepth]

Here's the typical strategy for your type of logic:

1) New FSD beta version releases, improves some static object response, doesn't improve others
2) Ignore things that were improved, point out things that weren't
3) Rinse and repeat until the logic no longer works and then be quiet

Here is the typical strategy for your type of logic.

1. You have no logic to your comments. (Tesla is 5+ years ahead but they still try to hit stationary objects)
2. Personal insults (always antagonizing whatever diplomat posts)
3. You never take any initiative to look up information that you are asking for (you ask for information that you can easily find yourself)
4. Always Tesla >>>>>>>>>>>>>>>>>>> everyone else.
5. Rinse and repeat until the next goal post.

 

[Bitdepth]

When powertoold states it is a minor issue, he means it is relatively easy to fix. I tend to agree. Yes, it is a major issue in terms of safety but that is not the point he is trying to make. He is saying the issues he sees are easy to fix, thus minor.

Tesla has already declared that HW3 will be insufficient for robotaxi.

It is not an easy fix, it literally involves every facet of the autonomous stack. You can work on it until you minimize the chances of you hitting an object but it is not going away.

Localization: know where you are in the environment so you don't run into objects
Perception: identify whats in your environment so you know where you can drive and where you can't
Prediction: understand what other agents are going to do so you can avoid running into them
Control: plot the optimum path to avoid hitting objects to get to your intended destination

 

[Terminator857]

It is not an easy fix, it literally involves every facet of the autonomous stack. You can work on it until you minimize the chances of you hitting an object but it is not going away.

Localization: know where you are in the environment so you don't run into objects
Perception: identify whats in your environment so you know where you can drive and where you can't
Prediction: understand what other agents are going to do so you can avoid running into them
Control: plot the optimum path to avoid hitting objects to get to your intended destination

Yes and Tesla has been working on those for 5 years. How difficult will it be to improve the map to say, don't driver there?
Perception has two ways of tackling this: 1) add those examples to the training data. 2) Voxel / bit depth.
Prediction: Not involved in a non moving pole / lamp.
Control: That software is done, but can be improved. In terms of avoiding the pole, just need to know that is an area to avoid.

 

[MP3Mike]

Tesla has already declared that HW3 will be insufficient for robotaxi.

Really? Where/when?

 

[powertoold]

Really? Where/when?

‘When Elon said he’s confident hw3 will get to 2-3x human safety during Autonomy Day. Tesla is aiming for 10x human safety to deploy robotaxis (I think).

 

[EVNow]

‘When Elon said he’s confident hw3 will get to 2-3x human safety during Autonomy Day. Tesla is aiming for 10x human safety to deploy robotaxis (I think).

That's a lot of tea leaves reading.

 

[powertoold]

That's a lot of tea leaves reading.

Yes, but considering how loose Elon is when gauging human safety, that's how I interpret it. For example, he talks about NoA being better than humans for highway driving already, and Tesla has some AP safety stats to show that AP augments human safety. Yet NoA is still level 2 and is nowhere "superhuman". So when Elon says he's expecting 2-3x for HW3, there's no way (IMO) they'd deploy a robotaxi with that, unless it's strategically geofenced (possible).

https://twitter.com/x/status/1387892143792984065

 

[Terminator857]

Really? Where/when?

Can't seem to find where Elon said this. It was after A.I. day.

 

[Daniel in SD]

‘When Elon said he’s confident hw3 will get to 2-3x human safety during Autonomy Day. Tesla is aiming for 10x human safety to deploy robotaxis (I think).

Not sure he ever said that but it should be obvious. HW4 probably only costs a few thousand dollars at most, it will pay for itself many times over in reduced liability when used in a robotaxi.
Here's where he talks about HW4 at AI day. "4 times more capable". Presumably it will also have much better cameras.

 

[EVNow]

Not sure he ever said that but it should be obvious. HW4 probably only costs a few thousand dollars at most, it will pay for itself many times over in reduced liability when used in a robotaxi.
Here's where he talks about HW4 at AI day. "4 times more capable". Presumably it will also have much better cameras.

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.

 

[powertoold]

Not sure he ever said that but it should be obvious.

Lol he literally said that in the clip you posted. "I think hw3 will achieve human safety 200-300% better..." (I'm paraphrasing)

I guess you were referring to my point about the 10x for robotaxis. That 10x part comes from the DMV exchange.

As for hw4 being 3-4x more capable, there has been a misinterpretation going on here:

Hw4 being 3-4x more *capable* doesn't mean it is 3-4x better human safety vs hw3... It could mean anything, but it's unrelated to human safety percentage. 3-4x more capable could lead to 20x human safety statistics, who knows. Or it could turn out Tesla never achieves human level safety with hw3, and hw4 will only be 1x human safety...

 

[Buckminster]

Wow - look how much progress Tesla has made in the past year. I have't seen a Waymo video for a while - they must be doing amazing things now - assuming their pace go innovation is similar??...