"While this is a step in the right direction, let's be real: it hardly solves the problem of robotaxis interfering with emergency vehicles. Asking a police officer, firefighter, or EMT to sit in the driver's seat and take control of a vehicle during an emergency is a Band-Aid fix."
That was my first thought as well. That feels like it would take way too much time in an emergency. I suppose it's better than nothing?
I have no idea what an actual solution would be though. Some kind of remote control / kill switch thingy? Everything I can think of seems kind of half baked at best.
In general, we should probably get all cars off the road until we can guarantee that they won't get into accidents. Once we have a 100% guarantee that no human lives are lost and no environmental harm can occur, we can slowly introduce one or two cars and see what happens. Perhaps the people who drive these cars should be highly trained professional drivers.
Then we can slowly introduce one or two amateur drivers and see if they cause any harm when going about their daily lives.
After we study the effects for a few decades we should be able to safely allow a few more people to drive cars on the road.
There is system in place for dealing with human drivers who misbehave in traffic - if it doesn't kill them, we can at least throw them in jail to think about what they've done.
Perhaps if robotaxi company executives, employees and scientists were liable when their cars messed up, they might be a bit less cavalier about letting them screw up traffic.
No one gets killed for screwing up traffic haha. And where Cruise operates, no one goes to jail for screwing up traffic. What does happen is people kill humans, and no amount of jailing the killers brings anyone back.
So let's make human driven cars safe before we let them on the road. When AVs kill as many people as human driven cars, we can switch.
Having an ability for a remote driver to quickly take over the vehicle via some kind of telepresence solution is one way to cover a percentage of the situations where these vehicles block emergency responders. But how to quickly initiate that? reliably?
If they cant interact with emergency vehicles well, which they cannot, then they shouldnt be on the road. This is only one of the growing number of obvious reasons why...
Would you also propose driving license suspensions for drivers that fail to move out of the way for emergency vehicles or illegally double-parked a vehicle and blocked passage of an emergency vehicle?
Perhaps because in the absence of the kind of alternative route to getting to work that autonomous cars offer, no one in (the US at least) is going to levy an economic death sentence for something that serious?
I'm so tired of this tail-wagging-the-dog excuse. It's so hard to operate in (parts of) our society without a car that we can't meaningfully hold drivers responsible for their bad behavior, or keep people who show a repeated disregard for their responsibility in operating dangerous machinery off the road?
Why is it a false dichotomy to hold software products to the same standard humans are held to? That seems less like a dichotomy of any kind and more like a consistent policy.
We generally do not send people to jail for failure to yield to an emergency vehicle. We do ticket and fine them, which you can also do to automated vehicles.
I’m not the commenter above but, something must be done about serial lawbreaking drivers. I see maybe 100 cars per day run stop signs, red lights, speeding excessively, and other dangerous driving actions with no consequences.
There needs to be some enforcement and yes I would suggest suspension if it is found that the behavior is habitual for a driver.
One of the scenarios in the article was the responder trying to wave the car back while the car was very insistent that it really needed to run over these fire hoses right now.
A motorist wouldn't just be suspended, they would go to jail and have their car impounded. Actually, maybe it should be tow operators and not first responders with surprise driver privileges. We'll have 100% compliance real quick if those are the guys watching out.
Here's the video from the article: https://vimeo.com/822783539. If a human driver was doing these things, the only reason it's not a "lose your license" situation is because the driver would have been removed from the vehicle by the officers and the vehicle moved aside by the officers. It's likely the driver still faces punitive measures. This is only loosely permitted because there isn't an individual in the car to blame.
The proposed solution is "We'll let you move the car." This is a reasonable "lose your business license" situation. It's pretty easy to sympathize with those who are "vandalizing robotaxis with blunt tools".
How long until a TikTok comes out with a video on how to impersonate a first responder needing to override a car, but it needs to be relocated to the liquor store/gas station/bank needing to be robbed first?
I think a lot of this could be automated with an exclusion zone around any emergency vehicle stopped within traffic space. Do not approach, leave if possible and always ask the operators for help if it finds itself within such an exclusion zone.
Cruise is level 4 pretending to be level 5. It needs a better understanding of what it doesn't know--and any emergency scene above a traffic stop appears to be beyond it's understanding. The robocars are too reliant on expecting other cars to behave that they fail with the cars that are allowed to misbehave and they don't understand that doing something "illegal" might very well be the right behavior. (I've done so when I was stopped at a light and sirens were coming up behind me. I could safely move out of the way but violating multiple traffic laws in doing so--any human would understand what I was up to but a robocar wouldn't even consider what I did.)
(Yes, my approach would mean that if a cop pulls someone over but remains in a traffic lane the robocar gets stuck behind it. I don't consider this a serious problem--that lane isn't moving anyway.)
Both Waymo and Cruise are L4 vehicles. Neither has claimed to be L5. Zoox has a vehicle model named "L5" that operates at L4 though. L5 would have an effectively infinite ODD.
Both seem to have claimed that L4 is Good Enough for city driving but there is mounting evidence to the contrary. Notably, L5 is the one that says, "drive the vehicle under all conditions" (https://www.sae.org/blog/sae-j3016-update). Apparently, "emergency response not underway" is part of the required conditions for their current implementation of L4.
ODD is about the design domain, not the actual domain of functionality. Not achieving the proper response to an emergency vehicle is simply a bug unless emergency vehicles were specifically excluded from the intended design domain for some reason.
I find it ... we'll say, weird ... where they've chosen to trial these vehicles already.
Is there some reason other than "we were able to get permits there more easily[0]" that I'm missing? I wasn't in Phoenix long enough to speak intelligently about that city, but I've done Austin and San Francisco on a few occasions in the distant past.
Both are high-traffic-with-pedestrians, much of it high-density and frequently changing traffic patterns ... except that most of them are varying degrees of "hellish" a whole lot of the time. Meanwhile you (likely) have ongoing maintenance of (likely many) high-traffic roads, detours, squeeze-around-work-crews all competing with emergency services.
Considering GM's HQ, I'm surprised they didn't elect to work these through Metro Detroit. We have all of that craziness at lower (and pretty predictable) traffic levels mixed in with roads that barely qualify as paved and some which are perfectly paved except for the tire-sized 2ft chasm that nearly borders the other tire-sized 2ft chasm. Figure out how to get these things to perfect the "pot-hole video game", understand the "Michigan Left that's also an Intersection[1]" survive more than a few months of trips from Metro Airport to "anywhere in Northern Macomb County" without dings, punctures or busted shocks then move to something more challenging. And we're a bit more tolerant of weirdness around "car sh!t" around here, especially since many in my area have jobs directly supported by the local autos.
I did notice that none of those cities are well known for large amounts of snowfall ... still, I must be missing something because it seems like there'd be much less disruptive places to refine your product. In places like these, I'd be handing the emergency vehicles a computer with a few buttons -- one of those things gets in the way, it's "Disable It" "GTFO" "Force it to Pull Over" maybe with fallback of a Joystick/Computer device to shove it out of the way. It almost sounds like they are providing some way that an emergency vehicle can control these things "on request" the way that reads. At least with a "three button/fallback joystick", mixed with "a little more adequate sensing/driving" work would make them no worse than a driver with the radio too loud who isn't checking their rear-view and would give them more power over the equivalent "machine driven" car than they have over the "d!ckhead driven" car.
Curious if anyone knows why, specifically, those cities were the test-beds?
[0] And one is also our HQ, which might be the only one that I can understand an exception (a desire to be close to your implementation solving the unique problems of that region that you probably understand better than most).
[1] This is a case where "traffic is U-turning to make a left" but the end of the U-Turn meets up with a road where traffic is turning right onto the road. In this case, "the law" is that "left hand turners yield the right of way" just like at any other intersection, but the vast* majority of time, the left-hand turning traffic mistakes their "green light" for a "green arrow" and assumes the usual rules of "go when it's green". It's so bad that they put signs* at many of these intersections reminding left-hand turners to yield on green. Follow the law, likely get hit. Follow the flow, and you'll likely get stuck there when you have the right of way. On a bad day, punch it with a hand on the horn if they move. :o)
Except the moment you get in the driver's seat of a car and especially operate it, you open a whole can of worms surrounding personal liability.
If I were an emergency responder in such a situation I wouldn't go anywhere near the interior of a robotaxi, let alone one in an exceptional circumstance with bystanders/potentially injured living things in vicinity or laying on the ground nearby. Fuck. That.
1 Is there a fine levied against the robotaxi operator when one of their cars interfere with an emergency vehicle?
2 Would it be feasible to put an emergency button in the back seat for the passenger (assuming car is not empty) to use in the case that an emergency vehicle is near and the car has not yet responded?
3 is it feasible for emergency vehicles to use their existing radio band to announce their presence to robotaxis without infering with voice transmissions?
I've had several really negative interactions with Cruise on San Francisco streets. More worryingly when I tried to bring them to Cruise's attention I got no response. I've had cars drive aggressively in a stop start motion towards me and my kids on a crosswalk, such that I had to stop in the middle of a busy street where we had right of way, to allow the Cruise vehicle to barrel past. A situation that would have resulted in being pretty dam pissed at a human driver. I don't think they should be on the roads.
51 comments
[ 3.3 ms ] story [ 107 ms ] threadI have no idea what an actual solution would be though. Some kind of remote control / kill switch thingy? Everything I can think of seems kind of half baked at best.
Take these robocars off the roads until they're read to respond to emergency vehicles properly.
Then we can slowly introduce one or two amateur drivers and see if they cause any harm when going about their daily lives.
After we study the effects for a few decades we should be able to safely allow a few more people to drive cars on the road.
Perhaps if robotaxi company executives, employees and scientists were liable when their cars messed up, they might be a bit less cavalier about letting them screw up traffic.
So let's make human driven cars safe before we let them on the road. When AVs kill as many people as human driven cars, we can switch.
There needs to be some enforcement and yes I would suggest suspension if it is found that the behavior is habitual for a driver.
A motorist wouldn't just be suspended, they would go to jail and have their car impounded. Actually, maybe it should be tow operators and not first responders with surprise driver privileges. We'll have 100% compliance real quick if those are the guys watching out.
The proposed solution is "We'll let you move the car." This is a reasonable "lose your business license" situation. It's pretty easy to sympathize with those who are "vandalizing robotaxis with blunt tools".
Cruise is level 4 pretending to be level 5. It needs a better understanding of what it doesn't know--and any emergency scene above a traffic stop appears to be beyond it's understanding. The robocars are too reliant on expecting other cars to behave that they fail with the cars that are allowed to misbehave and they don't understand that doing something "illegal" might very well be the right behavior. (I've done so when I was stopped at a light and sirens were coming up behind me. I could safely move out of the way but violating multiple traffic laws in doing so--any human would understand what I was up to but a robocar wouldn't even consider what I did.)
(Yes, my approach would mean that if a cop pulls someone over but remains in a traffic lane the robocar gets stuck behind it. I don't consider this a serious problem--that lane isn't moving anyway.)
Is this allowed?
Both seem to have claimed that L4 is Good Enough for city driving but there is mounting evidence to the contrary. Notably, L5 is the one that says, "drive the vehicle under all conditions" (https://www.sae.org/blog/sae-j3016-update). Apparently, "emergency response not underway" is part of the required conditions for their current implementation of L4.
Is there some reason other than "we were able to get permits there more easily[0]" that I'm missing? I wasn't in Phoenix long enough to speak intelligently about that city, but I've done Austin and San Francisco on a few occasions in the distant past.
Both are high-traffic-with-pedestrians, much of it high-density and frequently changing traffic patterns ... except that most of them are varying degrees of "hellish" a whole lot of the time. Meanwhile you (likely) have ongoing maintenance of (likely many) high-traffic roads, detours, squeeze-around-work-crews all competing with emergency services.
Considering GM's HQ, I'm surprised they didn't elect to work these through Metro Detroit. We have all of that craziness at lower (and pretty predictable) traffic levels mixed in with roads that barely qualify as paved and some which are perfectly paved except for the tire-sized 2ft chasm that nearly borders the other tire-sized 2ft chasm. Figure out how to get these things to perfect the "pot-hole video game", understand the "Michigan Left that's also an Intersection[1]" survive more than a few months of trips from Metro Airport to "anywhere in Northern Macomb County" without dings, punctures or busted shocks then move to something more challenging. And we're a bit more tolerant of weirdness around "car sh!t" around here, especially since many in my area have jobs directly supported by the local autos.
I did notice that none of those cities are well known for large amounts of snowfall ... still, I must be missing something because it seems like there'd be much less disruptive places to refine your product. In places like these, I'd be handing the emergency vehicles a computer with a few buttons -- one of those things gets in the way, it's "Disable It" "GTFO" "Force it to Pull Over" maybe with fallback of a Joystick/Computer device to shove it out of the way. It almost sounds like they are providing some way that an emergency vehicle can control these things "on request" the way that reads. At least with a "three button/fallback joystick", mixed with "a little more adequate sensing/driving" work would make them no worse than a driver with the radio too loud who isn't checking their rear-view and would give them more power over the equivalent "machine driven" car than they have over the "d!ckhead driven" car.
Curious if anyone knows why, specifically, those cities were the test-beds?
[0] And one is also our HQ, which might be the only one that I can understand an exception (a desire to be close to your implementation solving the unique problems of that region that you probably understand better than most).
[1] This is a case where "traffic is U-turning to make a left" but the end of the U-Turn meets up with a road where traffic is turning right onto the road. In this case, "the law" is that "left hand turners yield the right of way" just like at any other intersection, but the vast* majority of time, the left-hand turning traffic mistakes their "green light" for a "green arrow" and assumes the usual rules of "go when it's green". It's so bad that they put signs* at many of these intersections reminding left-hand turners to yield on green. Follow the law, likely get hit. Follow the flow, and you'll likely get stuck there when you have the right of way. On a bad day, punch it with a hand on the horn if they move. :o)
If I were an emergency responder in such a situation I wouldn't go anywhere near the interior of a robotaxi, let alone one in an exceptional circumstance with bystanders/potentially injured living things in vicinity or laying on the ground nearby. Fuck. That.
Some quality opining there...
2 Would it be feasible to put an emergency button in the back seat for the passenger (assuming car is not empty) to use in the case that an emergency vehicle is near and the car has not yet responded?
3 is it feasible for emergency vehicles to use their existing radio band to announce their presence to robotaxis without infering with voice transmissions?