Autonomous Car Risk to TSLA

[electracity]

No, not a Mod S driving into a school yard and chasing down screaming children.

But rather the risk that it seems certain that Tesla won't be the first to have an autonomous car used by civilians. The slow speed urban taxi, the Google Car, is the doable subset of full autonomous driving. What happens to TSLA valuation when the herds searching for the new, new thing shifts the "Goggle Car"?

Is an EV with a steering wheel the next generation? Or is it the last chapter of the old way of transportation?

Watching the 1/2 video posted early this morning of a v7.0 drive, I was struck with the enormity of the auto drive problem. I was struck with the thought, as a software developer, I would much rather approach the problem from a low speed vehicle in a defined environment.

As an investor, I would like a chart of the antacid and benzodiazepine consumption of the Tesla autonomous software development team. That video freaked me out.

 

[larmor]

Autonomous Car Risk to Insurance company

Start to short insurance car companies...

No, not a Mod S driving into a school yard and chasing down screaming children.

But rather the risk that it seems certain that Tesla won't be the first to have an autonomous car used by civilians. The slow speed urban taxi, the Google Car, is the doable subset of full autonomous driving. What happens to TSLA valuation when the herds searching for the new, new thing shifts the "Goggle Car"?

Is an EV with a steering wheel the next generation? Or is it the last chapter of the old way of transportation?

Watching the 1/2 video posted early this morning of a v7.0 drive, I was struck with the enormity of the auto drive problem. I was struck with the thought, as a software developer, I would much rather approach the problem from a low speed vehicle in a defined environment.

As an investor, I would like a chart of the antacid and benzodiazepine consumption of the Tesla autonomous software development team. That video freaked me out.

 

[jaguar36]

A Level 5, fully autonomous car is the point at which things will really change, and the point at which the Google car is trying to get to. I doubt Google will be the first one to get there though, its just to far away at this point to call. Best case is still at least 10 years, probably more like 30 years. Keep in mind that its not going to be the engineering that holds things up, but the politicians and the lawyers.

In short, its not going to have any effect on TSLA for the forseeable future.

 

[Quant]

No, not a Mod S driving into a school yard and chasing down screaming children.

But rather the risk that it seems certain that Tesla won't be the first to have an autonomous car used by civilians. The slow speed urban taxi, the Google Car, is the doable subset of full autonomous driving. What happens to TSLA valuation when the herds searching for the new, new thing shifts the "Goggle Car"?

Is an EV with a steering wheel the next generation? Or is it the last chapter of the old way of transportation?

Watching the 1/2 video posted early this morning of a v7.0 drive, I was struck with the enormity of the auto drive problem. I was struck with the thought, as a software developer, I would much rather approach the problem from a low speed vehicle in a defined environment.

As an investor, I would like a chart of the antacid and benzodiazepine consumption of the Tesla autonomous software development team. That video freaked me out.

I am am sure they have thought long and hard about this !

I expect no more than 10 k or 25 % will opt in and will pay the $2,500 for the auto driving software, initially. So, about 10 k drivers. The absolute risk therefore in the next 6 -12 months should be manageable. Also, the software may help avoid 1-2 of total accident that may otherwise have happened ? Don't know!

So, net net, it may be a better starting point now with just 10 am or so enabled, than later, for Tesla to learn and modify/improve.

Also, this version (on the highways ) is not a heck of a lot more risk, quantitatively speaking, than each of the several million cars with cruise control enabled, at any given point on today's highways.

And, I expect initial 80 % of more of early adopters will be vigilant , simply because it will take time getting used to. The mind takes time to adjust and trust.

they have to start somewhere.....

 

[electracity]

I think google is almost ready to go.

1) The dudes currently riding around in the cars become remote proctors via video
2) A private beta could move google employees between home and work.
3) A public beta would occur in a southwestern small city (no snow/ little rain or fog). Google may map the heck out of the area in 3D. They will know every block.

They could do the first million miles with a proctor watching every car mile. Then the proctors could start monitoring multiple cars. Later I assume the Google car will notify a call center in india when confused. As experience is gained, max speed is increased incrementally.

Google and similar is done with the big autonomous problems that need to be solved to implement. Tesla and other traditional car makers don't know where their implementation problems end. When you can't see the edges of the problem, there is no way to predict when the problem is satisfactorily solved.

For these reason I think Apple is doing a google car, not a Tesla competitor.

- - - Updated - - -

I am am sure they have thought long and hard about this !

Tesla didn't select their starting point for autonomous car. They decided to build EV, knowing autonomous would be in the future.

Google, and probably Apple, have proceeded from a starting point that allows delivery in a reasonable time frame. Silicon Valley entrepreneurship is all about delivering, not spending decades in research. This ethos can be seeing in Musk forcing autopilot out the door.

 

[liuping]

I expect no more than 10 k or 25 % will opt in and will pay the $2,500 for the auto driving software, initially. So, about 10 k drivers.

I think way more than 25% of the people ordering Model S's buy the AP convenience package. TACC is a very popular option. I would think it's more than 50%.

 

[dakh]

I think way more than 25% of the people ordering Model S's buy the AP convenience package. TACC is a very popular option. I would think it's more than 50%.

Yes the value of being able to read or watch TV while sitting in traffic is a lot more than $2.5K for most people. And it's pretty clear that the system is going to get to the point when it's safe to do so very quickly.

I voiced my skepticism here before on Tesla pursuing auto driving, but I am a bit more optimistic now. 50 engineers trying to solve this problem is not a lot, especially compared to the kind of resources others are throwing at this problem. But there are two mitigating factors. First, with a fleet of cars out there doing some real world driving they can now iterate very quickly and have immediate feedback. So I expect the system to get very good pretty quickly, within the limits of hardware it runs on. Second, it is quite likely that Tesla really hired the best of the best and indeed solutions to problems like this are moved along much quicker in a smaller team of top level people.

 

[adiggs]

A bit of a sidebar to this conversation, but at least in the US, we drive so many miles each year that driving statistics are typically standardized to 100 million miles driven. The numbers could be reported based on millions of miles driven, but a lot of those metrics would be fractions (2 orders of magnitude will do that for you).

It's relevant here in this way - I don't know how many miles will need to be driven in a beta program to start getting meaningful data that will move the needle with regulators, but 1 million miles isn't even a downpayment. 100 million vehicle miles means you have 1 unit of data to work with using the typical denominator used to report and understanding driving statistics (at least fatalities).


There is a LOT of driving and testing to come.

Here is 1 source with driving statistics:
https://en.wikipedia.org/wiki/List_of_motor_vehicle_deaths_in_U.S._by_year

 

[dakh]

Actually to the point above, Tesla is in the best position to make a case for the regulators of anyone so far. They can literally compare miles traveled under computer control vs. under human control, and also how many times humans had to interfere with computer control. They might just code up a feedback loop that allows the driver to press a "I had to intervene because autopilot made a mistake" button. And because of the iterative nature of this project, I think they're in a very good position to both improve the product and make a case for the regulators.

 

[omarsultan]

I would expect a major accident with AP at fault is an inevitability--much like a battery fire. And, much like the car fires, I think we need to cycle through it happening, TSLA dropping, media hysteria, media getting bored because its not all that common, TSLA rebounding so we can get on with our lives.

 

[BornToFly]

I am am sure they have thought long and hard about this !

I expect no more than 10 k or 25 % will opt in and will pay the $2,500 for the auto driving software, initially. So, about 10 k drivers. The absolute risk therefore in the next 6 -12 months should be manageable. Also, the software may help avoid 1-2 of total accident that may otherwise have happened ? Don't know!

Auto pilot is free for anyone that has the tech package, which I would think 80-90% of buyers already have.

 

[electracity]

What happens when autopilot is 90% better than today? How will people stay alert? This is going to be a very interesting experiment.

We already have a partial built-in autopilot, which is why we can drive, eat and talk at the same time. Tesla's autopilot will likely disengage our innate autopilot, once we gain sufficient comfort with the car's ability.

I'm wondering if human engineering is being minimized because Tesla has to be on the cutting edge.

 

[MikeC]

What happens when autopilot is 90% better than today? How will people stay alert? This is going to be a very interesting experiment.

We already have a partial built-in autopilot, which is why we can drive, eat and talk at the same time. Tesla's autopilot will likely disengage our innate autopilot, once we gain sufficient comfort with the car's ability.

I'm wondering if human engineering is being minimized because Tesla has to be on the cutting edge.

Some people always fear progress. Maybe if we put seat belts and air bags in every car, people will feel so safe and secure they won't drive safely anymore.

There was already the case of a guy who crashed who was incorrectly relying on TACC to stop in traffic. Tesla made it very clear that responsibility rests with the driver. I think most people understand that and accept it.

 

[austinEV]

What happens when autopilot is 90% better than today? How will people stay alert? This is going to be a very interesting experiment.

We already have a partial built-in autopilot, which is why we can drive, eat and talk at the same time. Tesla's autopilot will likely disengage our innate autopilot, once we gain sufficient comfort with the car's ability.

I'm wondering if human engineering is being minimized because Tesla has to be on the cutting edge.

I think it is really just similar to cruise control. I take my foot of the pedal and "watch and get ready to freak out mode" is engaged. Same with the wheel now. I think drivers will not disengage, not in this iteration. I think there will be moments where you feel like you needed to intervene (true or not) that will keep you from picking up a book. The next generation however... Auto driving natives will be getting into the backseats and napping.

 

[electracity]

I think it is really just similar to cruise control. I take my foot of the pedal and "watch and get ready to freak out mode" is engaged. Same with the wheel now. I think drivers will not disengage, not in this iteration. I think there will be moments where you feel like you needed to intervene (true or not) that will keep you from picking up a book. The next generation however... Auto driving natives will be getting into the backseats and napping.

I think most of us have seen people reading and watching TV while driving. Autopilot should be an improvement for those people, as well as drunks.

We know that many companies are puttering around California publically testing low speed driverless cars. It doesn't seem that Tesla is doing much of this type of development. From the driving videos it also doesn't appear that Tesla has delivered anything that many companies are incapable of releasing.

I get how companies will move from the low speed subset of autonomous driving to full implementation. I'm not getting how Tesla bypasses methodical development and comes out ahead in the end.

When a google car does something weird, the human monitor is busy noting what occurred in the environment. The developers get the sensor data and the notes on the event. Tesla isn't getting that quality of feedback from customers on autopilot. I'm sure that Tesla is running some driverless model S, but they don't have a lot of permits to do a lot of cars.

Tesla's developers likely do gain something, besides sleepless night, letting customers use this software. But I have a hard time identifying specific advantages.

 

[dakh]

It's been touched on during the call. One example was if on a particular part of the road drivers consistently take over and correct autopilot (which Tesla's cloud can easily detect), they can drive over there and test what happens for themselves. I'm sure there are other ways as well.

 

[electracity]

It's been touched on during the call. One example was if on a particular part of the road drivers consistently take over and correct autopilot (which Tesla's cloud can easily detect), they can drive over there and test what happens for themselves. I'm sure there are other ways as well.

Using Google street view is more likely

 

[dakh]

One more thing is worth mentioning: there's definitely incremental value in making isolated scenarios automated. The full-on transportation system overhaul due to self-driving vehicles is still in the future. Short to medium term though, here's a very distinct competitive advantage in selling a vehicle that, say, can auto-drive in traffic on the highway only. That alone would generate crazy amounts of sales/revenue.

For example, the implementation could simply sense that the car is on a part of the road (highway) where it's been proven to work 10x better than human drivers, safety-wise, and whatever other parameters it needs to be within (say, no ice/snow/rain in the first iteration), and light up a green light somewhere that indicates you can stop paying attention, and then alert with sound when you're close to your exit or leaving the area where auto-driving is supported.

 

[replicant]

Cool demo of the current AP system. Much more insightful than the media video of the last few days:

 

[tftf]

I don't understand why Tesla didn't at least limit these autopilot features to highways (using GPS checks etc.) - - as intended for the time being.

Nissan did this back in 2007 already for the GT-R: Nissan GT-R Detects When Car is on a Race Track, Disables Speed Limiter via GPS

It's only a matter of time before there will be accidents by people engaging this feature on roads not intended for autopilot use imho.