I agree that hardware 2.0 is unlikely to ever receive regulatory approval in western countries for level 4/5. Which is why I said "if". But all projections are probabilities. It is possible Musk will reach his goal.
But I also believe that most Tesla owners will not rent out their cars . The first time the car is returned with the smell of B.O. and mystery stains the dream of monetizing a personal vehicle will sour.
Cars used by strangers will likely remain corporate owned.
Autonomous driving is not a hardware problem, it is a computer science problem. It will be know to be solved when it is proven to be solved.
Tesla has a gazillion issues to solve before level four is possible. Talking about level five in existing cars seems to be a joke. Have they eliminated the mechanical brake linkage in the S/X starting this week? If they haven't, these cars are not level 5 capable.
What Musk is playing at is Uber's valuation. If Uber is worth more than $60B, then Tesla is too. (I do think Tesla is well ahead of Uber, but no one knows the distance to the finish line)
Autonomous driving is a solved problem. Has been for years. Google has racked up a lot of miles on their autonomous fleet, and they're just one of many groups involved. To make autonomous driving acceptable to the masses and convince the lawmakers to write laws to regulate it sensibly, you must defeat the FUD that it might kill people (because humans are TERRIBLE at assessing risk rationally), you have to prove that it is safer than humans over huge numbers of miles in all types of conditions. The only way to do that is to record a bunch of data. It would take decades to rack up the data required in a small fleet of vehicles owned by the company pushing the technology (see: Google). Tesla's adding the hardware to the whole fleet will rack up the miles at a rate orders of magnitude faster than Google could ever dream of.
The DARPA Grand Challenge started the ball rolling in many ways back in 2004. That first year, none of the autonomous vehicles involved made the finish line, and DARPAs million dollar prize went unclaimed. In 2005, they did it again, and 5/23 finished, and all but one surpassed the distance travelled by the furthest travelling car in the 2004 event. Since then, DARPA has run urban challenges and other events, consistently pushing the technology further and further. In 2006, I saw one of those vehicles that was trying to gain entry into the 2005 DARPA challenge and failed. In that era, these vehicles had numerous sensors on the exterior, and the bulk of the vehicle's interior was dedicated to housing a large amount of computing hardware. The pricetag for most of these vehicles was well into the mid 6 figure territory, as most of them used LIDAR units which were in the six digit range on their own. In the decade plus since, we've seen many of the required sensors get smaller and cheaper, and the software get better at using sensors intelligently, and the computing power required drop, as well as get more efficient packaging.
Autonomous driving is not an unsolved problem. The unsolved part is packaging the sensor suite and computing power in a way that is cheap enough to mass produce, and unobtrusive enough that it doesn't look ridiculous. Tesla seems to have done that with the AP2.0 hardware. It would be insane for regulators to not approve it when presented with the data. Eventually the people will revolt.
Fancy math for people interested:
Using this data:
Fatality Facts
Taking the Deaths per 100M Vehicle Miles Travelled by state as my population, I compute a standard deviation of 0.297 and it shows the overall mean at 1.08. Standard normal distribution suggests that 99.7% of all results will land within 3 standard deviations of the mean. That means the expected range of Deaths per 100M VMT is 0.189 to 1.971.
As of Elon's 7 Oct tweet, AP had logged 222M miles, and has 1 confirmed death so far. That's a rate of 0.450 Deaths per 100M VMT, or greater than 2 standard deviations below the mean, suggesting less than a 5% chance that happens from random chance, and a rate ~25% lower than any individual state. DC, RI, VT, and MN are the 4 lowest rate places, in the mid 0.6x range.
How many miles should AP have to prove itself better than humans over before regulators should approve it? Its already better than 2 standard deviations below the mean, but the sample size is small at 222M miles. The average reporting state has 59B miles these statistics are computed over. The smallest is DC at 3.5B miles. Standard deviation is 61B miles, so I could see an argument that perhaps you should prove it over (59B+(61B*3))=242B mi, and that its better than 0.189 deaths per 100M VMT. That would be unquestionably better (3 std dev above mean miles calculated over, 3 std dev below mean deaths per 100M VMT).