Holy shit I hate STEMlords sometimes
"just use AI to supervise the AI"
"why don't they just let it get stuck or crash like in murica?"
Everyone knows the superior system results in entire blocks being occupied by robotaxi traffic jams because of one strategically placed cone
Why are these idiots talking about Tesla here? Tesla autopilot is class 2 ADAS, compared to Baidu's class 4. There is miles of difference between the two. Tesla autopilot works entirely differently because it does very little mapping and very faulty sensors, meaning it tries to work everywhere but does so poorly. The serious players in this field such as Baidu are all using very precise mapping and LIDAR systems and only work on mapped roads. These are not comparable systems, nor is Tesla a serious player in this field that needs any attention whatsoever.
I still to this day laugh at the fact that tesla shipped lidar in their cars for years in anticipation of self driving features, but then just removed it later and went camera-only. I think mechanics were even instructed to remove it from existing vehicles that came in at one point.
Not a serious company.
You can tell that musk has never even played around with a Coral or done any work with vision. I do not trust the system that classifies me as a refrigerator and a sleeping bag as a blahaj in a life-critical operation. The only way I'd trust cameras, despite what the
would say is if every vehicle had an April tag on every corner. Even then, April tags, which are infinitely better that classifiers are known for shitting themselves and going insane twice a minute unless you do a Kalman filter with less trust in vision that 
's children have in him. Cameras fucking suck but I have an unhealthy relationship with them and I love them.yeah, even having only played around at an elementary level with apriltags, using mostly other people's pre-vetted code and hardware, 100% agree.
Good thing these aren't the type of people that are making decisions about self driving cars in the US... Oh wait
Waymo also uses human supervisisors very frequently. This is standard operating procedure during development.
oh yeah, its for sure pretty normal.
The implication is that they're behind technologically here (idk could be true but who cares, less bazingamobiles probably better, and they're being far more open than US tech firms which like to play "don't look behind the curtain" and pretend there's no humans involved when there actually are tons), and that for some reason these american stemlords don't think baidu, a $30 billion company, will ever get past having human supervisors for every car. That assumption might not be due to massive racism, but it really seems like the most likely reason to me.
