comma.ai: announces stopsign/red-light 'AP' - needs 10 cm GPS loc

[tander]

I wouldn't be dismissive at all. If you look what they've done with "off the shelf" cars with no custom anything is very impressive. I don't see this ever becoming more then lvl2 since they only have whatever the car offers, but still for what they have and what they have delivered it's quite good.

It also, makes you get that sick stomach feeling when he points out based on his knowledge and the Tesla current set of hardware that he believes Tesla will only get to lvl3 with it's current suite. I forget which video it was, but that was definitely eye opening, because if you watch his scopes, he's very pro Tesla and the Tesla approach.

Yeah I'm not super techy but for a handful of people in a house I'm impressed with what they are doing.

I don't remember hearing that about Tesla, in the video I saw he basically said google will be first, then Tesla and others with Tesla dominating on price. As I understand it Tesla can likely do level 4 now, it's just a question of regulation/insurance, and whether it is x times safer than the average human.

At first I thought that even 2x safer than the average human they should make it mandatory on all cars. But then I started thinking if I understand the statistics correctly, most drivers are much safer than the "average" statistical driver because the average is brought down by a small percentage or really statistically "bad" drivers. So for most drivers 2x safer than average might not actually be safer than what they currently are. It would be really interesting to see all the data Tesla has on it. I recall a talk by the guy that was head of autopilot from a year or two ago and at that time they were aiming for about 10x safer than the avg driver after 6 billion miles driven, wonder how that compares presently.

 

[BigD0g]

Tesla can do level 4 now???? @tander.

He basically says Tesla can win, because it doesn't cost them anything additional to add to the autonomous fleet and owners are footing the bill for the cars, so no cost for Tesla really. Whereas google wants to add another new car to the fleet it costs google 150k or so for the next car.

 

[tander]

Tesla can do level 4 now???? @tander.

He basically says Tesla can win, because it doesn't cost them anything additional to add to the autonomous fleet and owners are footing the bill for the cars, so no cost for Tesla really. Whereas google wants to add another new car to the fleet it costs google 150k or so for the next car.

Thanks I should have clarified, I just meant that internally I wouldn't be surprised if Tesla can do level 4 now, but of course not pushed to the public, and maybe only as safer or a little safer than the avg driver. The former autopilot head Sterling Anderson said they needed 6 billion fleet miles to achieve 10x safety and iirc that was about 2 years ago. That may have been incorrect but if true it seems like a reasonable guesstimate that that 6 billion is reached or approaching within the next year. Still have to wait for legislation/insurance though I think.

 

[Allanf]

@BigD0g - are you HW2? Elon has said they could upgrade HW2 computers if necessary.
Tesla Says Autopilot Hardware News Coming Soon

 

[Uncle Paul]

In the not too distant future, I would imaging that self driving road sensors would be added to our roads. Stop signs could have a little sensor mounted on top that would let auto pilot vehicles know they should stop there. Same with traffic signals, speed limit sign.

This would help the self driving vehicles drive precisely.

 

[BigD0g]

@BigD0g - are you HW2? Elon has said they could upgrade HW2 computers if necessary.
Tesla Says Autopilot Hardware News Coming Soon

Yes, I’m aware, we’ll see

 

[croman]

Yes, I’m aware, we’ll see

3 months, 6 months definitely. Jim Keller. AMD. Jim Keller. Tesla AI.

 

[BigD0g]

3 months, 6 months definitely. Jim Keller. AMD. Jim Keller. Tesla AI.

Let’s just go ahead and call that aPH3 speculation for now, or do we think they role out an Xavier chip before Jim’s chips are ready creating aPH3 and aph4 speculation, nothing in the code yet...

 

[croman]

Let’s just go ahead and call that aPH3 speculation for now, or do we think they role out an Xavier chip before Jim’s chips are ready creating aPH3 and aph4 speculation, nothing in the code yet...

Xavier only benefits nvidia. I don't see Tesla doing anything that expensive.

 

[BigD0g]

Xavier only benefits nvidia. I don't see Tesla doing anything that expensive.

Hmms, I disagree on this one near 4x the processing power at half the power usage, Tesla can certainly benefit from the added headroom. I’m sure a lot of work is spent optimizing the NN for the smaller processing power they have to work with. But we don’t know how far Jim is, so it’s really hard to say. I’d imagine they have to have some form f actual solution to optimize and tape out Jim’s hardware, until they have that they are just throwing crap at a wall and seeing what sticks, so Xavier might but them some nice headroom to work with.

 

[scottf200]

Below via TesLetter at Tesletter - Tesla news curated with love and delivered to your inbox weekly

22:32 -

TRAIN AI 2018 - Building the Software 2.0 Stack, by Andrej Karpathy

Tesla related stuff in this talk starts in minute 15:30. Some interesting bits:

• Since he joined - 11 months ago - the neural network is taking over the AP code base
• Tesla has created a ton of tooling for the people who tag images so they can be more efficient
• Around minute 20 he shows why labeling something that seemed easy like lane lines isn't as easy as it seems
• At 22:30 he shows traffic lights, some are really crazy!
• He talks about how random data collection doesn't work for Tesla. For instance, if you want to identify when a car changes lanes and you collect images at random most often than not blinkers are going to be off
• He mentions how auto wiper and how the dataset is crazy and it 'mostly works' 25:40