Re:Tesla's “full self driving” beta is potentially dangerous
by jillesvangurp on Friday March 19, 2021 @08:39PM
(#26519678)
That's actually going to be the key argument for Tesla (or any solution) leveling up from driver assist level 2 to full autonomy (level 3 to 6): it will have to be evidence based. The difference is that Tesla is actively accumulating a lot of evidence in the form millions of vehicles actively using this and gathering a lot of telemetry, video footage, etc. in the process.
As soon as it's going to be legal somewhere to remove your hands from the wheel, there will be incidents for sure. And then there will be responses to those incidents. At some point, that drops to acceptable levels and more will follow. Tesla will simply be able to pull out the numbers and say: look we're doing it here and it's fine. Solving the chicken egg problem of having the data to say that is key to ever getting to that point.
Tesla's strategy is not to lock itself up in an ivory tower for a few decades and to then release the perfect system. Companies following that strategy will fail. Instead, they are releasing imperfect software now and they are improving it rapidly and incrementally. As long as they have hands on the steering wheel (level 2) it does not actually have to be perfect. It will give them a steady stream of issues to fix. Lots of real world video footage to analyze. And a track record of slowly improving safety relative to human drivers in terms of numbers of incidents per million miles. Rolling this out to millions of drivers is going to get them a lot of data to work with very quickly. Data is the key thing to have in this space. Without data you can't make progress. Tesla is setting themselves up to get a lot of that. The reason they are getting cocky is that they already have good metrics.
It would be interesting to repeat the same journey with the same car in about a year. I bet it will still have issues but probably less of them. It's going to take more than two years for sure. But ultimately leveling up to 3 and 4 is going to have to be evidence based. And Tesla is building that case by gathering data from real drivers intervening when the software messes up and fixing things as that happens. That's something you can measure. The better that metric gets, the closer they get to their goal. Who else is even close to doing that? I think some of the Chinese manufacturers have shamelessly copied Tesla's strategy. And in some cases the actual software apparently. Regardless, China is the market to watch when it comes to full autonomy. A bit of wild west but they are moving fast there. IMHO if Tesla gets there, those manufacturers won't be very far behind. Unlike some traditional manufacturers who seem to be a bit hands off (bad pun) when it comes to moving forward on this front.
As soon as it's going to be legal somewhere to remove your hands from the wheel, there will be incidents for sure. And then there will be responses to those incidents. At some point, that drops to acceptable levels and more will follow. Tesla will simply be able to pull out the numbers and say: look we're doing it here and it's fine. Solving the chicken egg problem of having the data to say that is key to ever getting to that point.
Tesla's strategy is not to lock itself up in an ivory tower for a few decades and to then release the perfect system. Companies following that strategy will fail. Instead, they are releasing imperfect software now and they are improving it rapidly and incrementally. As long as they have hands on the steering wheel (level 2) it does not actually have to be perfect. It will give them a steady stream of issues to fix. Lots of real world video footage to analyze. And a track record of slowly improving safety relative to human drivers in terms of numbers of incidents per million miles. Rolling this out to millions of drivers is going to get them a lot of data to work with very quickly. Data is the key thing to have in this space. Without data you can't make progress. Tesla is setting themselves up to get a lot of that. The reason they are getting cocky is that they already have good metrics.
It would be interesting to repeat the same journey with the same car in about a year. I bet it will still have issues but probably less of them. It's going to take more than two years for sure. But ultimately leveling up to 3 and 4 is going to have to be evidence based. And Tesla is building that case by gathering data from real drivers intervening when the software messes up and fixing things as that happens. That's something you can measure. The better that metric gets, the closer they get to their goal. Who else is even close to doing that? I think some of the Chinese manufacturers have shamelessly copied Tesla's strategy. And in some cases the actual software apparently. Regardless, China is the market to watch when it comes to full autonomy. A bit of wild west but they are moving fast there. IMHO if Tesla gets there, those manufacturers won't be very far behind. Unlike some traditional manufacturers who seem to be a bit hands off (bad pun) when it comes to moving forward on this front.