Autonomous vehicles (AVs) — Waymos, in particular — are trundling their way around cities like San Francisco, Austin, and Phoenix like so many overly cautious teenage drivers. While they're well on their way to becoming just another piece of sorcery rendered unremarkable by ubiquity, there will be bumps along the road.1
Assuming the operational model remains a vehicle-for-hire, AV operators will have to deal with all of the messiness that comes with having human beings as customers. Those challenges will not only slow down the march to scale, they’ll also impact the design of AVs themselves over the long run.
Story Time
Once upon a time, I worked at Lyft.2 My first role at the company was on the customer service team answering emails and manning the 24/7 live support phone line…from 9pm to 5am. This meant I had peak levels of exposure to everything that can go wrong when you put a bunch of strangers together in a car.
People fight. They throw up. There was that one time someone defecated in the backseat of a driver’s car (we never figured out who). There was also the repeated scenario wherein a passenger would be sober enough to enter a vehicle, but not sober enough to exit (at least not under their own power). This one I actually experienced myself as a Lyft driver.3
It was late one Friday night and I was meandering around San Francisco picking up fairs. I received a ride request from a bar in the Mission District and when I got there, the passenger opened the door to the backseat, stumbled into my car...and promptly blacked out.
Now, this was back in the Stone Age, so he’d been able to request the ride without providing a drop-off location in the app. So, there I was with a passed out bro in my backseat and no idea where the guy was even trying to go to start working on his hangover.
Eventually he regained enough consciousness to kinda crawl out of my car and head back towards the bar, so all's well that ends well, but I go off easy. Those situation could be, and often were, much more difficult to navigate. Back in my role as customer service phone operator, I talked drivers through issues like this and worse five days a week. People are wild.
So it begins
So, people — especially under the influence of mind altering substances and in the semi-private environment of a moving vehicle — aren’t always their best selves. And this is already playing out in Waymos.
The original post (wayback page) from this reddit thread (comments still up) recounts the harrowing experience of a Waymo user opening the door to their ride only to find the interior slathered in ketchup.
From the post:
I showed friend Waymo for the first time. As soon as I got inside the car, I noticed ketchup smeared all over the display, steering wheel, and even inside the cracked eggs. (sic) There were also eggs on the side of the interior panel, with ketchup dripping outside. Called support and reported it, they ended the trip and said they would take care of it.
Knew this would eventually happen when LA went public. :/

The thread itself makes for good reading, too, because it quickly turns into trouble shooting. Folks do the natural thing everyone does and start thinking about what policies Waymo needs in place, what penalties they ought to apply, and what monitoring needs to be done to ensure enforcement.
Someone even asks the obvious question:
...are there any inward facing cameras ?
The answer is, yes, of course. It’s just that cameras are less of a deterrent than you might think. There are, however, even better examples to illustrate that point.
As reported in the San Francisco Chronicle, people are totally having sex in robotaxis. And I can’t say I find any of that too surprising given the things that I personally know people are willing to do even when there’s an actual human in the driver’s seat. A quote from one of the mobile-sex-enjoyers is illustrative:
I mean, there’s no one to tell you, ‘You can’t do that’
And that’s kinda the point. Knowing that someone could be monitoring your behavior is not the same as feeling like someone actually is; for better or worse, a human driver in the front seat will always feel more real than a Hal 9000 camera lens (even with the potential for a fine).
Necessity is the mother of product design
People are messy and cameras don’t guarantee bad things won’t happen, so where does that leave an AV operator? Well, after relating some of my experiences to transportation expert
, his immediate response was “modularity” (and I agree).We’ll likely see AV interiors become more modular such that everything is easily removable. This will help operations teams needing to clean dirty interiors or replace damaged components. Cleaning vomit out from underneath the driver’s seat would be much easier if you could easily strip the interior of a cab and power wash the entire inside. Just imagine interior components that let an AV operations team disassemble the inside of a vehicle similar to how a NASCAR pit crew handles the outside of a race car and that’s where I think we’re heading.
There’ll probably be some shift towards upholstery that’s easier to clean as well. Vinyl perhaps? Admittedly, I’m not furniture-materials-knower, but I’m guessing the requirements will be comfortable, but also extremely easy to clean.
Ultimately, AVs for hire and personal vehicles are different products with different use cases. Overtime, those differences will become reflected in their interior designs.
Some of the substance related issues may be more complicated. If we remember the scenario where someone passes out (or has a debilitating medical event in transit) there’s a lot of work to (a) identify an incident is occurring, (b) correctly classify the problem, and (c) do something about it.
On existing services, a lot of this is solved by virtue of having a human driver. In my experience, some combination of material incentive and human decency is enough for drivers to make sure their passenger gets home safe. For AVs, I can imagine a passenger failing to leave the vehicle triggering an alert, causing an operations staffer to look at a live feed for the vehicle. I’d further imagine they’ll have some internal taxonomy of incident types and interventions they’re allowed to take.
Now, none of this is rocket science. It’s not technologically difficult to route an AV to hospital if you believe someone is having a heart attack. The work will be in coming up with all the relevant policies in the first place; and many of these policies will be in response to types of incidences that will only start happening as AVs achieve much greater levels of scale.
That said, it’ll be fine (mostly)
None of this is a gotcha. I’m sure AV operators like the folks at Waymo are already thinking ahead about all of this and I’d bet cold hard bitcoin that more than a few of their Trust & Safety folks were recruited directly from Uber and Lyft.
That said, even if we’re talking about incidents that happen .5% of the time, that could mean… * does math on back of envelope*... tens of thousands of rides a month where something goes wrong.
What’s more, if there’s some bad thing that happens once every 10 million rides in an AV, it likely hasn’t happened yet (Waymo has done 5 million rides, ever). Once Waymo, or some other operator, is doing hundreds of millions of rides (Uber does tens of millions a day), it’ll become a recurring event. A system’s failure states become qualitatively different at scale.
So, as AV’s continue to scale as vehicles-for-hire, their design will have to account for these low frequency, high (negative) impact events that may disrupt service or else constitute a dangerous situation for riders. That will require time and effort, and it’ll be another challenge that operators will have to address as they continue rolling AVs out throughout cities across the world.
We fly through the air in giant metal tubes.
For the non-United States knowers, Lyft is the #2 in on-demand ride hailing in the U.S behind Uber.
If I were less earnest, I might tell you that I did over 800 rides as a Lyft driver because I was so CuStomEr oBsESsED; in reality, I just needed the extra money.
Great piece. Just took my first Waymo out in SF. I wonder if part of the plan is to start getting people acclimated to Waymo while people can ride in Jaguars (such a nice ride), then slowly over time, the cars will get grosser, eventually people will pay more for "premium" cars that have been cleaned more frequently while others will pay less for a car with ketchup all over. Surveillance and security cost money -- I'm sure those charges will get passed on to users somehow.