Re:Waymo’s Hiring Binge
by ItsDeathball on Tuesday February 25, 2020 @07:47PM
(#22419218)
I've thought something similar for a while, and it's been long enough that I think the problem is the failure mode. It might be that getting a car to recognize when it's encountered an edge case and needs to kick it to a human is about as challenging as just getting the car to come to the right conclusion and drive itself anyway. Intuitively we think of the problem as what the car needs to do, but the main problem is ascertaining the state of the road accurately. In the autonomous crashes we've seen from Tesla and Uber, it looks like the cars never even noticed enough to have an ambiguous condition they could deactivate autopilot from. They just had a completely erroneous perception of the outside world.
And I think that's due to a "let's keep driving" bias. In Uber's crash, for example, it detected an object moving in the road, but didn't understand what it was. Any reasonable human driver would have quickly stopped until it was clear what was going on. But the Uber car just kept going at speed until it was sure that it was a pedestrian, at which point a collision was inevitable.
I think the right thing to do is to invert the bias. If the car isn't sure it understands what's going on and that it's safe to proceed, it should stop. In which case, development becomes a long series of teaching about rarer and rarer circumstances, so it needs less and less human input to understand the road.
In the case of the Uber crash in Tempe, the car did kick the decision over to a human operator. However, it did so 1.3 seconds before impact. This wasn't enough time for the human to assess the situation and decide what to do.
We're talking about two different things. Uber's approach was horrific in many ways, and I agree that's bad. I'm talking about a car that is advanced enough to drive and stop safely, but may sometimes not be sure what to do, especially in situations that require social cognition.
Re:Waymo’s Hiring Binge
by anaanymouse on Tuesday February 25, 2020 @07:39PM
(#22419161)
There are many places where cars have to go with no cellular data coverage, even inside major cities. Tunnels for example.
Starlink won't grant Internet coverage inside tunnels or underground, but it plans to blanket all of the US with Internet coverage making that far more possible than with terrestrial cellular data networks.
I'm sure Tesla has been keeping that in mind as their self-driving technology progresses.
Sure. And you'll note that Waymo doesn't go anywhere like that.
The easy solution to this idea is to have the guy inside the car (you know, the passenger) pick up in this case. He's right there - no latency problems. The difficulty is in correctly identifying an unresolved edge case with enough margin to brake safely and draw the attention of a distracted passenger.
Sure, but then you lose the self-driving illusion, which is a big part of what's being sold. By all means have that as a backup mode if it can't phone home. But trained remote operators will not only preserve the fiction, but they'll be much more effective and they'll be the source of a lot of good feedback for the dev teams.
Self-driving cars are classified into 6 SAE levels, 0-5. 5 is full autonomy, 4 is autonomy in limited conditions, 3 is self driving but there has to be a driver ready to take over.
Getting to 5 is really hard. It could take decades, because there are a zillion edge cases. But what if you could turn cars loose at the "can handle 95% of driving" level and then let it call a human to give guidance when the car is nervous or confused?
Obviously, the phone-a-friend approach can't work for things requiring quick action. But if you bias the car to just stop safely and wait for orders, I think it's doable. E.g., the car comes upon an accident where the lanes in one direction are blocked. Humans know to go around the accident by slowly and carefully sharing the remainder of the road. So perhaps the car stops before the obstruction and asks for help. A human looks at it and draws in the path it should follow to get back to normal road.
So I think remote "robot coaches" could plug the gap between levels 4 and 5. And maybe some of the level 3 gap, too. And from what I hear of the Waymo operation in Arizona, I suspect Google's doing something akin to this.