What the Tesla disengagement report reveals (so far) is: during "Autonomy Day" it gave demo rides to investors in vehicles that it doesn't consider to be autonomous, while simultaneously releasing a demo video of a vehicle that it does consider to be autonomous.
This raises some big questions:
-where does Tesla draw the line between "autonomous" and "not autonomous" capability?
-did it tell investors that their rides were "autonomous" or not?
-why didn't it make this distinction between the rides and video more clear at the time?
-where does Tesla draw the line between "autonomous" and "not autonomous" capability?
-did it tell investors that their rides were "autonomous" or not?
-why didn't it make this distinction between the rides and video more clear at the time?
It's also not clear how Tesla is doing any kind of advanced development or validation of its neural net algorithms without doing any on-road testing. It must be using a mix of track work, simulation work and "shadow mode," which appears to be very limited
It's one thing for Tesla to make the boldest claims in the sector while offering less evidence of its capabilities than everyone else (one video in suburban Pal Alto)... but there isn't even strong evidence that critical development and validation work is really happening.
At a time when relatively tech-savvy VCs are less inclined than ever to invest in AV tech, the fact that public markets have so much confidence in Tesla's "Full Self-Driving" development based on so little hard evidence strikes me as incredibly risky.
One possible explanation is that Tesla could be doing public road self-driving development work in states other than California. A lot of states that allow AVs still require proof of insurance, permits etc so public record requests could rule this out
In any case, using non-autonomous vehicles for investor rides while using an autonomous vehicle for the accompanying video isn't a good way to get people to give you the benefit of the doubt on this issue. Especially given this context: dailykanban.com/2017/02/ca-dmv…
Also, if Tesla were testing outside of California why wouldn't they just record their hype videos there and avoid getting caught in apparent deceptions when the CA disengagement report comes out, the way they were for both their 2016 and 2019 videos?
This @andyjayhawk take is broadly accurate: disengagements are not good measures of overall progress in autonomous drive development. As this thread hopefully shows, however, it is a good way to sniff out what I'll diplomatically term "outlier behavior." theverge.com/2020/2/26/2114…
@andyjayhawk So, this enterprising fellow has found tweets from three people who got demo drives at Autonomy Day, and they all describe it as a demo of "FSD" or a "fully autonomous" system. This doesn't prove Tesla represented it as such, but it makes you wonder...
@andyjayhawk If Tesla represented these demo rides as being "fully autonomous" or "full self-driving" to investors but didn't report them as autonomous miles to the CA DMV, that obviously raises some very serious questions about Tesla's permit compliance and investor disclosure practices.
