It makes you wonder about the specifics:
- Did the 1.5 workers assigned for each car mostly handle issues with the same cars?
- Was it a big random pool?
- Or did each worker have their geographic area with known issues ?
Maybe they could have solved context issues and possible latency issues by seating the workers in the cars, and for extra quick intervention speed put them in the driver's seat. Revolutionary. (Shamelessly stealing adam something's joke format about trains)
FWIW, Cruise CEO takes to the orange site to dispel myths and rumors
the cruise CEO doing damage control (and lying their ass off) on the orange site makes sense, because all the former remote safety drivers I’ve talked to (admittedly none at Cruise) have been in agreement that self driving is absolutely fucked and the cars are essentially being remote controlled a majority of the time. some of the companies are also having what we engineers call fucking obvious problems like the high end GPUs they’ve packed into the cars overheating constantly, or the self driving software crashing due to bugs, both of which are situations that require immediate safety driver intervention
also, have the self driving crowd at the orange site never driven near one of these things? my area is known for aggressive drivers, but everyone gives the shitty robot cars a very wide berth (plus every exterior surface on these things is covered in extremely expensive sensors, and even the smallest operational self driving company has more than enough funding to bury you in lawyers before they ever admit fault in an accident)
Fair enough, I will note he fails to specify the actual car to Remote Assistance operator ratio. Here's to hoping that the burstiness readiness staff is not paid pennies when on "stand-by".
it would be entirely consistent with what he is saying for the car to ask "should I run over this child" every few minutes, as long as it only takes the operator say six seconds to click NO
more generally, it would be consistent with his claims for someone to be almost constantly watching the video and inputting instructions, as long as the instructions are discrete and do not constitute taking sustained control of the vehicle. if you had a vehicle that had to be manually informed every time a light was to be stopped at or a lane was to be changed into, as long as you don't take over the wheel that would be completely hidden in his stats. he's 100% bullshitting
from my current understanding, this is extremely likely to be the case. it’s an incredibly unsafe system that doesn’t scale beyond small-scale testing, but this is most likely the only practical way to divide the attention of a relatively small number of safety drivers between multiple cars that need constant correction
given their struggles with GPU overheating (which’d eliminate the car’s autonomy entirely), I wonder how many remote safety driver systems are remote commands injected into an OEM driver assist (ACC/LKAS/automatic lane change) package, since those come with the car and are rather a lot more rugged against obvious shit like heat than the bullshit a bunch of fresh grads with zero automotive experience bolted to the car
It’s suspicious that the criticism is in terms of miles traveled and the response is in terms of time. My own car operates flawlessly 98% of the time in a driverless mode called “park.”
What is my purpose?
You fix 100+ traffic cone gridlocks per hour.
Oh my god.
