What happens if these cars get in a situation where there is no legal way to move? For instance, the other day I was driving down the street and it was closed because of a Farmers market. Will they just sit there for hours? In gridlock, there is often no way to move without "blocking the box."
Waymo has an operations center the car is in contact with. I would assume the car would call out to this operations center for human-provided instructions. And you'd be the only car to experience it, since all other cars would be notified and take an alternate route.
Part of getting this out into the world is finding good answers to these questions.
Waymo and Cruise automation are building out call centres, where remote human monitors will be available to help out the vehicles when they get hung-up. The monitors won't be able to make safety-critical interventions, but they'll be able to give the cars and passengers instructions, kind of like a backseat driver, I guess.
So if there's an accident scene, for example, the human monitor will judge whether to stop and wait, navigate a path through, or turn around and find a detour.
Ah, the cat is out of the proverbial bag! These cars are not really "self-driving" at all! They are merely remote-control cars with adaptive cruise and lane assist.
So it's somewhere between a Lyft and remote surgery. What happens if the car loses internet?
No, they're not remote control. The human operators can't make safety-critical interventions, they won't be able to drive a robotaxi 50 mph into a tree, for example. There's an inherent liability associated with that.
> Ah, the cat is out of the proverbial bag! These cars are not really "self-driving" at all!
?
Do you class a self driving vehicle as one without any manual controls whatsoever? In which case, you're going to be waiting a long time for your version of it, and not for tech reasons.
Presumably Waymo has this traffic data by leveraging existing Android monitoring technology and can reroute accordingly. If you've ever used Google Maps for directions and come up to an accident or something you know it will suggest a better route as it becomes available. These cars will likely not get themselves into high-congestion areas with no way out simply because they already know about the high congestion area and reroute accordingly
There are anecdotes of Google Maps causing traffic, routing people to dead ends, etc. There was the incident where Google Maps routed a lot of interstate drivers down a dirt road serving a prison. Last week in Oakland, Google routed drivers away from traffic on I-880, onto Embarcadero where the bridge is closed for construction. Last week, Google refused to route me over the open, clear Sonora Pass between Modesto and Bridgeport, giving me instead a 2.5-hour detour via US-50 to South Lake Tahoe.
I'm sure this Waymo project does not take the perfection of Google Maps data as a given.
I wasn't aware of these, but with anecdotes like these then it makes perfect sense to have available remote human overrides! I don't mean to insinuate that maps is perfect, just that the number of times the cars will find themselves in this specific situation is probably minimal.
Eh, in my whereabouts Google Maps routinely suggest "shortcuts" on shitty dirt roads instead of 10km "detour" on a perfect tarmac road. And use 70km/h (speed limit for dirt roads) to guesstimate duration for the shortcut. It got to a point that I'm double-checking their suggestions via sat photos or other routing services.
Also local human overrides. I as a passenger want to determine the route to take. Maybe I just don't like the recommended route, maybe I know that it won't work. There's no reason why the passenger shouldn't be able to determine the route.
> I'm sure this Waymo project does not take the perfection of Google Maps data as a given.
Agreed, it would be nuts if they did.
Construction happens all the time. One day there is a bridge, the next day there isn't. Roads get destroyed by storms. I'm sure they have some way of detecting that the map doesn't match reality and have a variety of ways of dealing with that.
>> If you've ever used Google Maps for directions and come up to an accident or something you know it will suggest a better route as it becomes available.
So, extremely poorly then? Maybe it's different in the US but Google maps is extremely poor around EU, with roads leading nowhere, zero knowledge about no-turn signs and one-way roads. Traffic only works in largest cities and even then poorly. My TomTom dedicated satnav beats it by every measure possible.
Perhaps Google Maps is poor in the EU, but it's clear that Google does know how to make good maps, as they do elsewhere. If that's necessary for autonomous vehicles, that's what they'll do.
I think this whole thing is hilariously premature for a bunch of other reasons, but I assume a remote piloted car is gonna go REALLY slow until the car is confident again, and Google's cars are ridiculously limited in their geographic area. My guess is that they have sent a Verizon guy ("Can you hear me now? Good.") down each and every road they whitelist the cars to drive on.
This is a major reason for Google street view cars going everywhere, I would guess. Letting people look at street view in google maps is not the reason they have and continue to spend millions (billions?) on creating these detailed maps. Sort of like when Google had a free info phone service for a while to train their voice recognition system. I would not be surprised if street view goes away at some point.
a lot of people drive with van full of networking gear, telco companies checking their coverage maps, telco companies checking the competition maps, regulators checking coverage, consumer magazines checking coverage maps, regulators checking something else but seeing the coverage while doing it, mapping cars, etc.
my points are 1) running the roads scouring for GSM and wifi networks is nothing out of the ordinary 2) they would probably not even have to create a mission for that, and use or piggyback something else.
Everybody else is suggesting remote human intervention from an operations center, but those would likely suffer from hold times and dead zones.
What if, as an additional option, a touchscreen in the car highlighted the car's planned route, and the humans inside could drag the line to reroute, and/or add waypoints to the route? This would get around most farmer's markets and road closures. And/or they could tap to indicate a semi-permanent obstruction, and the car would do the rerouting.
In New York, they sometimes block all of Lexington Ave. (a major north-south throughfare) at Midtown for pedestrian activities. What happens when a hundred self-driving cars all get to Lex and stop because they can't pass?
These aren't "wacky what ifs." I don't know what things are like in the Valley, but here on the east coast something unexpected changes my commuting route a couple of times a month (multiply that by tens of thousands of drivers).
"Do whatever the human would do, but more efficient" isn't an answer. These cars can't get on WTOP the day before and listen to the warning that the Pope is visiting and X, Y, Z streets will be blocked off. One can imagine various contingency scenarios, but the devil is in actually handling all the edge cases.
>"Do whatever the human would do, but more efficient" isn't an answer. These cars can't get on WTOP the day before and listen to the warning that the Pope is visiting and X, Y, Z streets will be blocked off. One can imagine various contingency scenarios, but the devil is in actually handling all the edge cases.
Google maps already accounts for changes in traffic patterns and actively detects road closures, both based on municipal announcements and traffic flow data.
So yeah, they'd work the exact same way that I commuted on the east coast (which is the same way I commute in SV): the maps application had traffic information updates faster than the radio.
The answer is "do whatever the humans do, but more efficient", because the same technology has already made humans more efficient.
> Google maps already accounts for changes in traffic patterns and actively detects road closures, both based on municipal announcements and traffic flow data.
Sure, but Maps did try to send me the wrong way down a one way street last week, and through a closed tunnel last night. It's excellent technology, I don't mean to mock, but it's not perfect - situational info is going to have to override map info.
I think the question is less whether self-driving systems will ever be right and more what they'll do when they (like humans) are inevitably wrong. Driving in major cities often means breaking normal traffic laws because construction, police guidance, or some other circumstance means there is no legal way to progress.
That's the sort of problem that puts L5 vastly further out than this. What does your car do when a traffic cop waves it through a red light, or someone rear-ends it and needs to meet you at a crash reporting zone half a mile down the highway?
> The answer is "do whatever the humans do, but more efficient", because the same technology has already made humans more efficient.
I don't really buy that. Google Maps regularly gets confused in D.C. due to changes in traffic patterns. And Apple Maps once tried to send me down a highway on-ramp that had been closed for several weeks. They also get confused at the beginning/end of trips ("proceed to the route"). I don't even try to catch Ubers from downtown D.C. anymore because mapping applications can't reliably get drivers to a pickup point without spending several minutes coordinating with the driver over the phone.
Where do you think that radio station is getting its info? I promise you that Dan in Chopper5 is not actually feeding them real-time updates. That same data source (state DoT and municipal traffic data) is already fed into most mapping apps and you can see it just as soon as someone at the station gets a traffic update and has to decide when to feed to update into the broadcast, but you don't have to wait until minutes ending with a 6 to get the data now.
> "Do whatever the human would do, but more efficient" isn't an answer. These cars can't get on WTOP the day before and listen to the warning that the Pope is visiting and X, Y, Z streets will be blocked off.
Well yes, they can. Planned closures like that are the easiest ones to handle. They are planned and broadcasted. They can simultaneously listen and account for all planned & broadcasted road closures. That's easy.
The harder ones are going to be things like after a storm with a bunch of trees knocked down and random obstructions all over the place.
And yet Google Maps has on multiple occasions failed to recognize when a on-ramp to a major California Interstate was closed (even after the road work had been announced days in advance in huge street signs). I was forced to set my navigation to Avoid Highways for a bit to get it to consider an alternative.
And don't get me started on Germany. You don't stand a chance if you attempt to use Google Maps in Germany to navigate around construction (which is basically always happening somewhere)--at least that's how it was two years ago when going through country roads, despite the presence of well marked closure and detour signs.
My point is, these may be the "easiest" types of things to handle, but we have a long way to go on execution. Getting something to work in one locality (e.g. Arizona) is a far cry from making something work worldwide.
It's too bad they can't rollout self driving cars in places with less extreme edge case scenarios like... Chandler AZ and then build up to more challenging scenarios like NYC!
The thing is these edge cases being given have relatively easy solutions and when consider in the context of Google has cars driving themselves they do not seem like a huge issue.
But the sooner we get to the next phase so sooner they can be better understood and solved. I want this to get rolling. It is time.
