> CNBC also asked Cruise for information about typical response time for remote operations, and how remote assistance workers at Cruise are trained.
> “More than 98% of sessions are answered within 3 seconds,” the spokesperson said.
> During that incident, Cruise previously told NBC, its vehicle “braked aggressively before impact and because it detected a collision” but then tried to pull over and in the process pulled the pedestrian forward about 20 feet.
> The CEO wrote, “Cruise AVs are being remotely assisted (RA) 2-4% of the time on average, in complex urban environments. This is low enough already that there isn’t a huge cost benefit to optimizing much further, especially given how useful it is to have humans review things in certain situations.”
So was the Cruise vehicle being remotely operated when it dragged the pedestrian 20 feet? (That's the impression I've had since this was initially reported)
The whole act of remotely operating automobiles on public streets strikes me as extremely dangerous, unaccountable, and that it'd be obviously prohibited by law.
AV advocates can at least argue the sensor suite provides far more data than a human driver has at their disposal. But a human driver on the other side of a wireless link working with a modicum of visual camera feedback and perhaps audio is unquestionably inferior to a licensed operator in the driver's seat. This is simply insanity unfolding on our city streets.
> The whole act of remotely operating automobiles on public streets strikes me as extremely dangerous, unaccountable, and that it'd be obviously prohibited by law.
I Googled it and that doesn't seem to be the case; in fact it is mandated by law:
> The "Driverless AV Passenger Service" pilot program allows for the provision of passenger service in test AVs without a driver in the vehicle. Under this pilot program, a communication link between passengers and "remote operators" of the vehicle must be available and maintained at all times during passenger service.
Which seems likely to me to be a relatively new legal status carved out specifically for autonomous vehicles.
In a pre-AV world, my impression has been that it's been absolutely illegal to remotely operate vehicles on public roads. Maybe I'm wrong though. My impression here has been influenced by admittedly bored and unnecessary neighborhood police harassing me in my youth for purportedly illegally operating RC cars on public streets, which didn't even weigh thousands of pounds.
It might be interesting to understand how much lobbying/$$ was necessary to get the AV laws on the books as written today.
If they have that many interventions, you have to imagine there are situations of needed split second interventions that aren't able to happen.
Since they rely on HD mapping, a lot of them are probably when the environment is altered by construction etc. and if that was an overwhelming majority and can be detected far in advance of an operator being assigned I think it would more ok, maybe that's the case? I'm not sure that could add up to such a large percentage unless they handle it as an exception over and over again across the fleet when there is construction.
You drive in sf a lot? There’s double parked cars every two blocks, trash everywhere on the street, random crazies in the middle of streets, etc. Not hard to imagine something unexpected happening every few miles. Mind you 5 miles is almost entire length of the city
> The CEO wrote, “Cruise AVs are being remotely assisted (RA) 2-4% of the time on average, in complex urban environments. This is low enough already that there isn’t a huge cost benefit to optimizing much further, especially given how useful it is to have humans review things in certain situations.”
That's a pretty worrying statement: CEO is of the opinion that decreasing it isn't possible and is trying to put a positive spin on it. Driving (pun not intended) remote assistance down should be their primary goal, to drive down OPEX and facilitate market expansion.
I think that what you're imagining is like a driver who can let the car do its thing and just intervene now and then once every fifteen or twenty minutes. I believe that what the article is describing is a guy in front of a bank of monitors watching 15-20 cars and intervening in one or another pretty much constantly for the duration of a work shift.
> As far as the ratio of remote assistance advisors to driverless vehicles on the road, the Cruise spokesperson said, “During driverless operations there was roughly 1 remote assistant agent for every 15-20 driverless AVs.”
I knew one company who mined the Twitter firehose for events to sell to news stations and the government who had over 60% of the events identified and surfaced by dedicated employees, and 100% of the AI identified events were vetted by a person before going out.
Still valuable, but it was always telling how those employees were never really acknowledged outside of the company.
Of course. Amazon Go stores are a good example of this as well. As are all kinds of content filters across the internet. A large percent of "AI" is cheap labor in third world countries sitting in front of a computer and clicking buttons. And if human assistance isn't happening in real time, there is still a ridiculous amount of it in the training stages.
A UK company called spinvox blew up when it was revealed that their speech transcription tech was actually a warehouse filled with people in Pakistan. In that case the blow up was due to data protection laws in the first instance, and investor confidence in the second instance. They had, of course, told everyone that their transcription system was a giant e-brain.
If most of the transit opportunities for Cruise is less than 5 miles in San Francisco, as most people are not going end to end or even end to half; wouldn't this mean it's not every ride that this will happen. Though it's still bad in the sense of it claiming to be self driving
https://news.ycombinator.com/item?id=38145997