How else to interpret 2016 CA autonomous driving results?

[-=buzz=-]

My take on an explanation/interpretation is this:

AFAIK Goggle/Waymo approach is extensive surround mapping, and a designed algorithm to navigate through this perceived world.
Tesla on the other hand, is using a Neural Network which is learning to drive by itself, based on the input of its sensors.

Therefore, a disengagement for Waymo indicates, that they encountered a situation their algorithm couldn't handle AT ALL. Google will have to extend their algorithm to handle this new situation correctly in the future.
In Teslas case, a disengagement means that the NN just learned not to take this line of action for the given situation EVER AGAIN (Or at least add a significant negative bias to this path). So each disengangement actually improves the AP.

As others have mentioned over time the disengagements should asymptotically approach 0.


If you now imagine all current AP2 cars running in shadow mode, and every beviation between AP and driver behaviour is used as an input for the Neural Net, this should give a massive amount of input from improvment over short periods.

Based on this input from all Tesla drivers, the AP should actually develop an averaged driving behaviour, which means that AP should drive better than 50% of the human drivers. And thats before adding always following traffic laws (no speeding), not getting tired, faster reaction times and having 360° vision.
It should even develop some kind of intuition for dangerous situations, based on drivers easing off the gas or tapping the brakes in certain situations..

 

[tomas]

Yeah, this whole thing is clearly a planned scam. Build the roadster, raise money, build the S, raise money, deliver superchargers, raise money, build gigafactory, raise money, introduce auto pilot, raise money, build X, raise money, add sensors and introduce AP2, raise money, reveal the 3.... then bugger off to Mars. Yeah... that's the plan!

Seriously. Why do conspiracy theorists always project the worst. Couldn't there be a conspiracy to do good??

 

[jldf310]

This is embrassing. Are you saying that tesla is testing on some top secret difficult road that google or even other companies don't have access to?

How can you be this naive? They have testing in the same cities. this is apples to apples.

I have to disagree on this one. I don't think OP is stating that Tesla is operating on secret difficult roads - I think it's been documented that Google has been operating in a fairly controlled environment. The suburban roads of Mountain View that Google has mainly operated on are much different than the streets of New York. I don't believe you can call it apples to apples.

 

[McRat]

I have to disagree on this one. I don't think OP is stating that Tesla is operating on secret difficult roads - I think it's been documented that Google has been operating in a fairly controlled environment. The suburban roads of Mountain View that Google has mainly operated on are much different than the streets of New York. I don't believe you can call it apples to apples.

Perhaps in the first 1/2 million miles I would agree. Look, EVERYBODY knows Google has a massive headstart. But nobody knows what they intend to do with the system. The wise bet would be to compete with Uber or license the technology out.

If they license it out, those companies who buy it will save billions and years of development.

If they Uber it? It will replace taxis globally. Trillion dollars a year?

 

[Phoenixhawk101]

I don't see anything interesting in this data. It simply shows that Google has a head start...if you didn't know that already you haven't been paying attention.

My issue is more that Gooogle hasn't done anything with their development. At this point it is still just a toy to show off with no real target to market.

 

[stages]

From what I have read, I believe that Google/GM/Tesla are all looking to out-Uber Uber. The real prize is changing the way it which we use cars and car ownership in general. I have advised my kids to hang on to their cars until about 2020 as self-driving cars may save them lots of money.

 

[Vitold]

It's simple really. Tesla had faked the disengagements in order to lull the competition...and give Tesla shorts opportunity to take part in a squeeze of a century.

 

[Bladerskb]

I have to disagree on this one. I don't think OP is stating that Tesla is operating on secret difficult roads - I think it's been documented that Google has been operating in a fairly controlled environment. The suburban roads of Mountain View that Google has mainly operated on are much different than the streets of New York. I don't believe you can call it apples to apples.

And Palo Alto is new york? haha please stop. Tesla tested on an easy road and had disengagements every 3 miles.


Besides Google tests in 4 cities;
Mountain View, Cali
Austin, Texas
Kirkland, Washington
Phoenix, Arizona

Also Cruise/GM released a video and they test in San Francisco and their stats are far far better than tesla.

Why can't you people accept the fact that googles self driving system as last publically tested in california is just not as good as some of the competition

 

[jeffro01]

View attachment 213013 View attachment 213012 I usually get a laugh out of the ability of fanboy culture on this site to rationalize almost any bad Tesla news. (Though lately it seems the messiah has left the building now that he is working cooperatively with the new administration!). But now, I come to you in my time of need.

Help me see the new results showing google averaging 5,000miles between driver intervention as anything other than clear indication that Tesla's AP2.0 bluster is anything more than hype...

(Tesla averaged 3miles per intervention and did a total of 550 miles vs Google 630,000)

What is the point of this thread? Seriously? Why are you here trolling...??? Do you not have anything else better to do than come here, insult the community, and then post some bogus chart which means almost nothing in a deliberate attempt to rile people up? Is your life really that devoid of happiness?

AP2 hasn't even fully been released yet and you're already ready to throw in the towel and call it nothing more than hype?

Why are you even here?

Jeff

 

[Bladerskb]

My take on an explanation/interpretation is this:

AFAIK Goggle/Waymo approach is extensive surround mapping, and a designed algorithm to navigate through this perceived world.
Tesla on the other hand, is using a Neural Network which is learning to drive by itself, based on the input of its sensors.

WRONG.

Google is using a combination of machine learning, deep learning neural network, reinforcement learning and combination of other techniques.

Therefore, a disengagement for Waymo indicates, that they encountered a situation their algorithm couldn't handle AT ALL. Google will have to extend their algorithm to handle this new situation correctly in the future.
In Teslas case, a disengagement means that the NN just learned not to take this line of action for the given situation EVER AGAIN (Or at least add a significant negative bias to this path). So each disengangement actually improves the AP.

If you now imagine all current AP2 cars running in shadow mode, and every beviation between AP and driver behaviour is used as an input for the Neural Net, this should give a massive amount of input from improvment over short periods.

This is laughable. This is not how neural networks work.
Second of all the models do not train while its being run. That is called reinforcement learning not deep neural network that tesla is using.
Your car doesn't learn jack nothing. All it does is run the models that it downloaded from tesla.

The Deep Convolutional Neural Networks are trained inhouse by tesla not in your cars. Your cars just run the models.
Shadow mode is simply validation mode. The variation isn't used for input for NN (because that's not how it works).
But for validating the latest model based on disengagements (aka a big difference from the system's prediction versus the ground truth of the driver's action).

 

[u00mem9]

What is the point of this thread? Seriously? Why are you here trolling...??? Do you not have anything else better to do than come here, insult the community, and then post some bogus chart which means almost nothing in a deliberate attempt to rile people up? Is your life really that devoid of happiness?

AP2 hasn't even fully been released yet and you're already ready to throw in the towel and call it nothing more than hype?

Why are you even here?

Jeff

Um...what? Reading half of the first post and then skipping to the end to post outrage is a nice confirmation of my tongue-in-cheek jab. Thanks for the help!

As you can see, this discussion is about what the publicly released data tells us about the progress at the companies working on autonomous vehicle technology. Since we have very little solid info about the progress of each, we have to speculate by comparing demo videos, leaked info, and now this data.

If that discussion makes you upset, feel free to check out some of the other threads available on this forum.

 

[jeffro01]

Um...what? Reading half of the first post and then skipping to the end to post outrage is a nice confirmation of my tongue-in-cheek jab. Thanks for the help!

As you can see, this discussion is about what the publicly released data tells us about the progress at the companies working on autonomous vehicle technology. Since we have very little solid info about the progress of each, we have to speculate by comparing demo videos, leaked info, and now this data.

If that discussion makes you upset, feel free to check out some of the other threads available on this forum.

BS... Pure and simple BS. You didn't post this to inform the community, the opening to your post confirmed that. You posted this to intentionally troll the community. You're out to dump on AP2 when it hasn't even been fully released yet...

You can claim all you want that you're just some innocent little snowflake coming here to start an honest discussion all you want. The reality is clearly far from that.

So I ask again, why are you even here?

Jeff

 

[u00mem9]

I'm here the same reason we all are. I own a model S. I am very interested in autonomous vehicle technology. I keep a close eye on what company looks like they will be first-to-market.

Last fall, I thought that was likely to be Tesla after the big announcement. Now, with this data, it looks to me like Tesla is way behind Google. I'm not sure any other way to interpret the results. You are welcome to offer your thoughts. You mind find the process therapeutic.

Basically the only OEM that I'm aware of signing up to apply Google tech is Honda, and I don't really want to be stuck in a front wheel drive appliance to get autonomous operation

 

[McRat]

I'm here the same reason we all are. I own a model S. I am very interested in autonomous vehicle technology. I keep a close eye on what company looks like they will be first-to-market.

Last fall, I thought that was likely to be Tesla after the big announcement. Now, with this data, it looks to me like Tesla is way behind Google. I'm not sure any other way to interpret the results. You are welcome to offer your thoughts. You mind find the process therapeutic.

Basically the only OEM that I'm aware of signing up to apply Google tech is Honda, and I don't really want to be stuck in a front wheel drive appliance to get autonomous operation

Why would you want a "driver's car" that you bought so you would not have to drive?

 

[u00mem9]

Why would you want a "driver's car" that you bought so you would not have to drive?

I don't really see them as mutually exclusive. Porsche 911 with the ability to wake me up when we arrive at the track day!

 

[Xminus6]

It seems quite obvious to me that the disengagements in that report are purely related to making the demo video (most of them also being in the rain), and are not at all indicative of the overall state of Tesla's development.

That's right. The clustering of results suggests it was around the production of that video. So, the disengagements might have been unrelated to the actual performance of the car and more about meticulously planning the route, and the production of the video. I have no reason to believe that Tesla's system, in October, was any way equal or superior to Waymo's system, but these stats are very hard to interpret equally.

 

[zambono]

If you actually look at the report that Tesla submitted and not just the overall summary you'll see that they only did testing in October (and a bit in Novemeber, aka right around the time they were making the demo video. Since then they have had ZERO testing miles.

So from this we can draw two conclusions. They only did it to make a fancy video and aren't doing any more development work and are basically just ripping everyone off with the FSD hardware which they don't ever intend on following through with.

Or.. They are doing all of the developmental testing somewhere else. Either on private roads, or in a different state. (or most likely, both)

It seems quite obvious to me that the disengagements in that report are purely related to making the demo video (most of them also being in the rain), and are not at all indicative of the overall state of Tesla's development.

What I said in the second post, not even knowing about Teslas report. The amount of miles is very low, so I assumed it was just based on creating the video.

 

[jldf310]

And Palo Alto is new york? haha please stop. Tesla tested on an easy road and had disengagements every 3 miles.


Besides Google tests in 4 cities;
Mountain View, Cali
Austin, Texas
Kirkland, Washington
Phoenix, Arizona

Also Cruise/GM released a video and they test in San Francisco and their stats are far far better than tesla.

Why can't you people accept the fact that googles self driving system as last publically tested in california is just not as good as some of the competition

It appears Google has good test lab results, maybe you should go buy one. Oh, wait . . .

 

[Phoenixhawk101]

Well Tesla responded to this so we no longer need to guess. They are doing autonomous testing, just not in Calinfornia so it doesn't need to be reported. They are not disclosing how many miles they have done or where they are testing.