Autonomous Car Risk to TSLA

[Ludus]

I think Google previously had the largest fleet on the road feeding data from the wild and that's about 50 vehicles. Within weeks Tesla will have thousands. Soon tens of thousands of vehicles constantly streaming data and being updated with new software iterations. This access to a serious flood of data for machine learning is a huge advantage to Tesla. I don't know what percentage of vehicles that are eligible will end up with autopilot but I suspect Tesla will arrange for it to be a solid majority. By the end of 2016 if there are 100K vehicles that implies about a Billion miles driven by autopilot equipt vehicles. How many vehicles will Google or anybody else have streaming data by then, a few hundred?

 

[electracity]

I think Google previously had the largest fleet on the road feeding data from the wild and that's about 50 vehicles. Within weeks Tesla will have thousands. Soon tens of thousands of vehicles constantly streaming data and being updated with new software iterations. This access to a serious flood of data for machine learning is a huge advantage to Tesla. I don't know what percentage of vehicles that are eligible will end up with autopilot but I suspect Tesla will arrange for it to be a solid majority. By the end of 2016 if there are 100K vehicles that implies about a Billion miles driven by autopilot equipt vehicles. How many vehicles will Google or anybody else have streaming data by then, a few hundred?

Tesla will able to track the locations where the autopilot frequently quits. At quitting the autopilot will show that there is not enough coherent data for the autopilot to proceed with confidence. While that data will be somewhat useful, it does not record the specific issue because the system, by the fact it is quitting, isn't getting adequate coherent data.

If cars were able to send a great deal of useful data back to Tesla the autopilot wouldn't have to quit. That's why everyone else has human monitors sitting in the car. Tesla does too, but not many according to permits.

Google car handling a four way stop with other cars and pedestrians is far advanced to what Tesla released with v7.0.

 

[dakh]

Tesla will able to track the locations where the autopilot frequently quits. At quitting the autopilot will show that there is not enough coherent data for the autopilot to proceed with confidence. While that data will be somewhat useful, it does not record the specific issue because the system, by the fact it is quitting, isn't getting adequate coherent data.

If cars were able to send a great deal of useful data back to Tesla the autopilot wouldn't have to quit. That's why everyone else has human monitors sitting in the car. Tesla does too, but not many according to permits.

Google car handling a four way stop with other cars and pedestrians is far advanced to what Tesla released with v7.0.

We don't know exactly how this works. I wouldn't be surprised if the car sends substantial amount of data back describing the signals that made it bail on auto driving. Kind of a core dump. And yes what Google does is more advanced. We don't know though if Tesla can also do that, just that they'd rather get something working out the door and get all the PR.

 

[electracity]

We don't know exactly how this works. I wouldn't be surprised if the car sends substantial amount of data back describing the signals that made it bail on auto driving. Kind of a core dump. And yes what Google does is more advanced. We don't know though if Tesla can also do that, just that they'd rather get something working out the door and get all the PR.

What seems to be missing is GPS/Nav integration. Perhaps the system will be more impressive with those inputs assisting the logic.

 

[gregincal]

Yes the Google version is far more advanced than Tesla, but also far more expensive. It probably won't be sold to individuals at all. It's a threat to the entire current way of buying cars, and thus a threat to Tesla. But no more that to any other car manufacturer.

 

[electracity]

We don't know what high quality Lidar costs in large volume. Private car utilization is ~4%. Car as a service will have much higher utilization. So the Google system may be cheaper per mile.

Here is Google saying Tesla's approach likely won't achieve full autonomous driving.

Chris Urmson: How a driverless car sees the road | TED Talk | TED.com