A Waymo drops the author off on Haight St. in San Francisco, in the middle of the lane in front of ... [+]
Driving people around safely is a super-hard problem, and has been the top priority of all self-driving software teams. But if you want to be a robotaxi, you can’t drive people around unless you pick them up and drop them off. This more mundane task wasn’t at the top of the priority list, and now that Waymo and Cruise are doing full robotaxi services in San Francisco, Phoenix and Austin, that task, called PuDo, is getting them in to trouble.
Waymo began in a suburb, where finding curb space to pull over is easy, and many destinations are places with parking lots. Some other companies run shuttles with fixed routes and official stops for doing PuDo. They won’t pick you up or drop you off just anywhere. In San Francisco, people expect service like a taxi.
Taxis, Uber and Lyft cheat on this all the time. They commonly stop for a short time in the traffic lane to handle PuDo, temporarily blocking traffic if there’s no easy place to pull over. They get away with it. The city of San Francisco has not been so forgiving, and in their recent letter to the California PUC about the permits Waymo and Cruise use to operate, they cited bad PuDo explicitly as a reason to slow down their expansion.
In Cruise’s permit application, they indicated that as a commercial vehicle they are permitted to briefly double park for PuDo under the vehicle code if no suitable curbside spot is available.
When Kyle Vogt, Cruise CEO, personally did the first robotaxi ride with Cruise in San Francisco, it was a bit surprising to see the vehicle just stop in the traffic lane. It was late at night with nobody around, but Vogt had actually summoned the car while standing in front of a large loading zone for a temple, a perfect place for his pick-up. He crossed the street to get the car — and even the other side of the street had several good PuDo spaces. The city saw this and complained, and Cruise and Waymo got better, but not perfect.
Turns out it’s hard. In fact, if used use Uber or Lyft, you probably have had problems trying to find your car, or for the driver to find you, and set the spot for pick-up. Drop-off is usually much easier since you are in the car and can say, “Drop me there...”
A recent video showed a couple summoning a Cruise. It drove by them, and as they approached it enter, it zipped off, then drove two blocks, including up a steep hill. They were out of breath when they got to it. It hunted around for a spot, probably because of the pressure from the city. It probably didn’t take the steep slope of the road into account.
These cars have, or should have, a map of all the semi-permanent PuDo spots in the city. That includes loading zones, but does not include bus stops, where it’s illegal. It does include any gap made by a too-short parking space between two driveways, as long as a motorcycle or mini-car doesn’t use it. In some, but not all places, fire lanes and other red curbs can be OK for PuDo as long as the vehicle doesn’t park. No parking zones can be great, but not no-stopping zones, of course. In places where fire lanes are not allowed, they should be enabled for PuDo by responsible companies who can promise they won’t use them if they hear sirens or a fire report is present for the area.
Riders can be shown all the nearby locations in the app. Riders should also be able to say they are in a spot which is currently clear, which is to say a parking spot with adjacent space, or if nothing else is near, a single parking spot. If their ride is near, they can hold that spot from being taken for a few minutes. The app should know the location of all parking spots in the city which their car can quickly enter without needing to back in. (Backing in is possible, but may take a fair bit more time.) Zoox vehicles, with their 4 wheel steering, will have an advantage here.
Waymo at the Superbowl
Waymo has been doing a lot of promotion around their service in Phoenix during the Superbowl. Their vehicles don’t go to the stadium, though, so they won’t tackle the really big problem of serving (eventually) tens of thousands of riders at once. The stadium there is not very accessible on transit, it’s mostly surrounded by huge parking lots. You could serve people with robotaxis waiting in those lots if you had enough of them. You could even coordinate them so they leave in waves for superior flow without stop-and go.
A better plan would be to have vans and buses, perhaps even human driven, that take groups of people to staging lots in all directions. You would be pointed to the van going the direction you are already going, and it would leave within minutes to take you to a lot with a set of robotaxis already there, and get in yours for the final leg to your destination. There’s almost no limit on the capacity of such a system if you have enough buses, and the robotaxis do shorter trips so you need fewer of them.
This weekend, though, Waymos are avoiding a large area of the downtown. If you book a trip into it, you are sent to special drop-off spots that aren’t particularly close to the destinations. It seems the Waymos may want to avoid doing PuDo in the most congested areas around the special venues. This may be wise in the early deployment of a robotaxi service, though an Uber will take you anywhere cars can drive — though there is debate about whether it should or not.
PuDo blocking the lane?
On top of this, however, it is reasonable to be as tolerant of PuDo in the street as we are with human-driven taxis. The key is to not do this when traffic is thick — which the vehicle can know, and the rider can advise — but also to require that the riders will enter the car extremely quickly.
Right now there is an issue. Many riders are unfamiliar with their robotaxi, and both companies have taken the step of requiring the rider to unlock the doors using the app on their phone. People take time to figure this out and press the button. This is different from the taxi, where the door is already open (or is unlocked by the driver as the rider approaches) and the passengers can slip in quickly. They can in fact do it more quickly than the traffic delay that would occur if the taxi tried to parallel park and pull out again, so it’s a win for everybody — if it’s sure to be fast.
Not all riders are that fast. Some may be less nimble, or they may have bags for the trunk. Riders can designate in the app that they will be quick when the time comes, and know to not designate that if they have bags for the trunk. Riders can in fact be timed on how well they pull off the pick-up, and be de-rated if they are too slow.
For this to work, the car needs better information on the location of the riders. GPS only goes so far. Riders can pin themselves on a map but then may move. If the car knows the Bluetooth MAC address of the rider’s phone, it can tell when the rider is very close to the car, which can signal to stop and to unlock, and possibly just which of the pedestrians in the area is the rider. The car could also temporarily get a copy of the rider’s face to do face recognition. A bit more complex the rider could even have a companion take a picture of them as they look that moment — a bit hard in a selfie — so the car can recognize them and how they are dressed. (When I call an Uber in such situations, I will tell them I’m the big guy in the blue shirt, and this is an analog of that.)
Another trick that can work is for the app to flash a blink code in the screen or LED of the phone. I often hold my phone out when I see my Uber/Lyft approach to tell them “I summoned you with this phone” while I point at them. This would be even better, and the car would quickly spot its rider who is blinking with the right pattern. The blinking might even be just in color or otherwise invisible to humans, depending on the phone. The key is that the car knows exactly who and where the rider is.
Drop-off is, of course, much easier. The rider has a touchscreen if they want to mark spots on the map, but in any event there is no need to match the rider and car, just a need to get dropped off as close as possible to the actual destination. It’s also easier to do fast exit.
Of course, cities and states can make rules on when “in the lane” PuDo is approved or tolerated. They can regulate it by hours, or dynamic traffic data. (Thanks to online tools, there is excellent data on where traffic is heavy.) Of course it would not be done if there are convenient out-of-lane spots. Because the rider would confirm that this will happen, they can also be given liability if they end up being slow and blocking traffic, rather than the robotaxi company. Indeed, if the city wants a fine for blocking traffic, the robotaxi company can charge the fine to the rider’s credit card — after lots of warning and agreement that they will pay a charge if their pick-up blocks traffic for more than the approved time.
Again, consider that the alternative might be for the car to stop, blocking the lane and back into a PuDo space. That blocks traffic too, though it does offer safer entry and exit to the curb. It’s the right thing if the traffic is not thin. If there is a nearby PuDo space that can be entered going forwards (ie. slightly longer than a usual parking space) then that should be used, unless the rider is disabled. Riders with disabilities that inhibit them from walking a long way to a PuDo stop may be granted an exception — even though they will block traffic even longer getting in right in front of their door.
