That is an interesting take on Autopilot.
- Constuction zones and temporary/changing lane markers.
- The trade-off between lane-following and lead vehicle-following gives some insight into the logic involved.
- Express lanepoles, whatever those are.
The Tesla roadmap has Autopilot in beta until it is
'10x safer' than humans.
Which it already is or not depending on how you muster the statistics. Lives saved (like crime prevented by guns) won't show up in the statistics, but there is already
anecdotal evidence. AEB ≠ Autopilot.
Quote:
|
Originally Posted by jamesqf
So they don't even actually know how their "autopilot" works, they're just throwing data at a big neural net, and hoping something useful sticks. I can just imagine what a liability lawyer could do
|
'Fleet learning', if it even employs neural nets, operates on the back end. The result would be delivered to individual Tesla's through a software update. If the lawyer[s] can penetrate that, and bring a jury along with them on the journey, then that's ...awesome?
The deep problem with fleet learning is privacy. If you decline to participate when you buy the car then features are disabled. But they don't know about those late-night runs to dispensaries just across the state line.