It is hard to think of a technology that has been more over-hyped than autonomously driven vehicles. Predictions from just a few years ago made it seem like self-driving cars would be commonplace by now, the ranks of Uber and Lyft drivers would be plummeting, and cities would be forever changed. After all, it was just 2019 when Elon Musk proclaimed at the Tesla Autonomy Day event that:
“By the middle of next year, we’ll have over a million Tesla cars on the road with full self-driving hardware, feature complete, at a reliability level that we would consider that no one needs to pay attention.”
“From our standpoint, if you fast forward a year, maybe a year and three months, but definitely next year for sure, we’ll have over a million robotaxis on the road.” 
Of course, Elon Musk is a showman with a history of over-promising and under-delivering, but others have been almost as outlandish – such as the 2016 prediction by Lyft president John Zimmer that the majority of trips by the ride-sharing company would happen in autonomous vehicles within 5 years. Or Business Insider predicting in 2016 that 10 million self-driving cars would be on the road within 5 years.  Even allowing some leeway for the hyperbole that normally accompanies a bold prediction, these statements were wildly over-optimistic.
What has gone wrong? It turns out that while the basics of driving a car are relatively straightforward, the nuances of driving safely in an urban environment are much more difficult than autonomous vehicle engineers anticipated. Is the person standing by the curb getting ready to step into the crosswalk or just waiting for someone to pick them up? Is the car that is slowing down and drifting slightly getting ready to turn or is the driver just distracted by their phone? Human drivers are amazingly good (but not perfect) at figuring out the intentions of other humans. Computers, not so much. Progress is being made, but it is time to reset our expectations about what is going to happen in the next several years.
To be fair, many new cars have implemented elements of self-driving technology but none are so feature complete that “no one needs to pay attention.” With one exception, there are only a handful of cars available for sale to the public where the driver can take their hands off the wheel and let the car drive itself. And even then, there are (1) severe limits on where self driving can be used, (2) the driver must always remain focused on the road, and (3) must always be ready to assume control over the vehicle.
The exception is the Mercedes-Benz EQS and S-class models that can self drive up to 40 miles per hour without the driver staying focused on the road. Although the driver can play games on their phone or answer emails, they must still be able to retake control if so directed by the car. This is a milestone of sorts, but still well short of what people have predicted.
In February of 2022, I posted an article that gave a broad overview of the autonomous vehicle industry. I stand by what I wrote at that time, except that I may have been slightly too optimistic about the pace of progress. Perhaps in five years I will look back at that article and hang my head in shame, but I think there is a decent chance that I won’t be too embarrassed. In this post, I’m going to give an update on the state of the autonomous vehicle industry with a particular focus on self-driving taxi services (or “robotaxis”) that have the potential to change urban transportation dramatically.
Taxi cabs, of course, have been around for decades but it was just 2009 when Uber was founded (then called UberCab) and it has only been in the last 10 years or so that Uber (and others) have become commonplace in their current form. The convenience of being able to move from point A to point B by simply using a smartphone app which matches available drivers to your desired trip has been more revolutionary for city life than most people realize. In 2022, Uber alone provided over 7 billion rides to a base of 130 million active users which generated gross bookings of $115 billion. 
Amazon Zoox Robotaxi
If that transportation innovation could be reduced in cost from the current $1 - $2 per mile average to perhaps 50 cents per mile, the impact would skyrocket. A robotaxi ride would suddenly be an affordable option for many more types of people in many more types of situations. That is the potential of robotaxis – by eliminating the driver the cost of travel (at least in theory) drops substantially. This would lead, in turn, to changes in the transportation choices that people make. Inexpensive robotaxi rides would likely reduce trips in privately owned cars, and might also impact the use of public transit.
A secondary benefit is that by eliminating the driver, there is now more room in the vehicle for passengers. In fact, nearly all of the major players in the robotaxi industry are planning on producing or buying vehicles specifically designed for this function. There won’t be a driver, of course, but there also won’t be a steering wheel or gas pedal. Purpose-built vehicles will not only have more passenger space, they will be less expensive than modifying a traditional car, be easier for users to enter and exit, and be cheaper to operate. Leading the way in the US is Amazon with their Zoox vehicle (currently entering testing), Google/Waymo with their ZEEKR platform, and GM/Cruise with their Origin vehicle.
I was in Phoenix recently for a short vacation and it gave me the opportunity to ride in a robotaxi driven without any human assistance. Waymo (a subsidiary of Google parent Alphabet) has been testing self-driving vehicles on public streets in the Phoenix metro area since 2017 and has been offering rides to the general public as a full-fledged rideshare service since 2020. In January of this year, Waymo reached the milestone of one million rider-only miles. I couldn’t wait to give it a try!
Like any rideshare service, the first step is to download an app onto your phone and create an account. The Waymo One app is easily installed and works more or less like any rideshare app. The robotaxi service is limited to two different parts of the Phoenix area, including the downtown district and Sky Harbor airport. I landed relatively early in the morning and decided to take Waymo from the airport to a restaurant (about 8 miles away) in order to get some breakfast. I admit to being a little nervous as I started the app and requested a ride, but my experience ended up being awesome and I wouldn’t hesitate to do it again.
My ride did, however, give me some insights into the realities of autonomous vehicles and some of the issues that robotaxi companies are facing as they try to become as ubiquitous as Uber or Lyft. At the time of my ride, there was no inclement weather and traffic was modest by big-city standards. If it had been raining or if traffic had been heavy, my experience might have been different and Waymo might have included an “autonomous specialist” to provide a human backup to the “Waymo Driver” computer system.
The trick to bringing an autonomous vehicle to market is not being able to drive in good weather and light traffic, it is being able to handle tough conditions and unforeseen circumstances – what the industry refers to as “edge cases.” Every time an edge case is encountered, the system “learns” something new, but human intervention is occasionally needed. In Waymo’s case, that is becoming increasingly rare but I suspect that there are still a lot of edge cases that haven’t fully been addressed, particularly in areas where bad weather is more common.
Overall Impressions. My ride in the Waymo vehicle was almost entirely uneventful (I will get to the exception in a moment). It is a lot like driving with someone who is extremely conscientious about following every single traffic law and best practice – in short, a bit boring. Undoubtedly, that is exactly the experience that Waymo wants users to have because boring also means safe and reliable. Although boring, it was never irritating. The car accelerated briskly, turned without hesitation, and stopped normally. It merged into traffic without any problems and never felt indecisive or unsure of how to react to other cars on the road.
It was a ride with the precision one would expect from a computer – it drove precisely the speed limit, it stayed precisely in the middle of its lane, it never tail-gated other vehicles, and it stopped precisely a car length behind the car in front. The Waymo Driver showed none of the human foibles that most of us have picked up over years of driving. For example, when turning left onto a multi-lane street, the vehicle always turned precisely into the inside lane – no lazy “banana” shaped turns for Waymo!
Pick Up and Drop Off. Anyone who has hailed a cab in a big city knows that while cabbies try their best to find a safe spot to pick up or drop off passengers, if necessary they will stop in the middle of traffic and endure the honks of irritated vehicles behind them. Uber and Lyft drivers seem a little more careful about the pick up/drop off process but will still make judgment calls that occasionally bend the law for the convenience of the passenger. Not Waymo.
In fact, the pick up/drop off process was distinctly different from most cab or Uber rides. To begin with, I was directed to take the Phoenix Sky Train from the terminal to the 44th and Washington Station which is a much quieter location than the area outside the terminal normally used by Uber and Lyft. This was not a difficult thing for me to do and my Waymo car (a modified Jaguar I-Pace) was patiently waiting for me when I arrived, but it did add five minutes to the process. My guess is that the traffic chaos outside the main terminal is an “edge case” that Waymo still has difficulty with. What was more surprising was that Waymo would not drop me off directly at the restaurant, but found a quiet street a block away where it could pull safely out of traffic. The restaurant had a small parking lot but it was an odd layout that the Waymo Driver apparently did not feel comfortable negotiating.
This pick up and drop off process (sometimes referred to as the “PuDo” issue) is something that cities will need to plan for in the future. I’m sure robotaxis will get better, but for safety reasons they are never likely to block traffic or double-park the way a human driver might. This means that cities may need to create loading zones on almost every block face, particularly in congested areas, in order to accommodate robotaxi traffic.
Edge Cases. My ride was not without a couple of quirks that probably would have qualified as edge cases at some point in the development of the self-driving software. First, we encountered a traffic backup due to some road construction. My Waymo vehicle detoured around the problem by cutting through a mixed commercial and residential area rather than stay stuck in traffic. It drove more slowly since it was on side streets, but I’m sure we saved time. We also passed a man walking his bike in the street on the wrong side of the road. The Waymo Driver stayed several feet clear and passed safely without incident.
Finally, there were a couple of instances in which I detected a slight “twitch” by the Waymo vehicle in response to other cars on the road. For example, an on-coming car suddenly turned left into a driveway when I think most drivers would have waited for us to pass. It wasn’t really a big deal but it was unexpected enough that I probably would have lifted my foot from the gas pedal and been at least ready to brake. The Waymo vehicle not only stopped accelerating but actually did start to break until it realized what the car was doing. My guess is that this is an example of the difference in reaction time between computers and humans. The computer was able to react so quickly that it felt unnatural to me, but the slight twitch that I felt was actually the system being safer than a human driver.
The Route. My 8-mile trip ended up taking about 25 minutes and the route consisted almost entirely of thoroughfares with a 35 mile-per-hour speed limit. The odd thing about the route is that it didn’t include any part of I-10, an interstate highway that cuts through downtown Phoenix and that I think would have saved me about 10 minutes of travel time. Perhaps I-10 was congested at the time, or maybe the Waymo vehicles stay off of highways as a matter of standard practice. I don’t know and I wasn’t able to find any information that would answer that question one way or the other.
It did make me wonder, however, whether cities would be smart to work with robotaxi firms to develop preferred streets that robotaxis would take whenever possible. For starters, cities could make sure that the preferred routes had lane markings in good condition and safe places for the pick up and drop off activities. But it might also be an opportunity to start implementing the next generation of the “internet of things'' technology. Specifically, the ability for pieces of roadway infrastructure to communicate directly with cars. A traffic signal, for example, might broadcast its timing sequence so that cars could slow down when a signal is about to turn red, or road sensors might broadcast the optimum speed of travel to minimize traffic congestion. This technology would not need to be limited to robotaxi companies, of course, but they might be in a financial position to be the first to take advantage of it and they might be willing to partner with cities to develop and test the systems needed to make it work.
The One Exception. Earlier I wrote that my trip was almost entirely uneventful. The exception had nothing to do with my Waymo vehicle being unsafe or inconvenient in any way. Instead, it had to do with the interaction between human drivers and the computerized Waymo Driver. As I mentioned, the Waymo vehicle obeys all traffic rules including the speed limit. This means that on the thoroughfares we used the Waymo vehicle was going slower than just about everyone else. As in most cities, drivers routinely go five to ten miles per hour above the speed limit especially when traffic is relatively light and the weather is good. There were two instances when other vehicles honked at us, presumably because they thought we were going too slow. This is a bit odd, of course, because honking has absolutely no impact on how the Waymo Driver behaves and you would think that Phoenix drivers would know that by now. The Waymo vehicles are decked out with so many sensors that they are almost impossible to mistake for a regular, human-driven car.
One incident stands out because it escalated to an unusual degree and because it illustrates that the combination of computer-driven and human-driven cars might not be all sweetness and light. We were on a four-lane thoroughfare in the inside lane driving exactly 35 miles per hour. A pickup truck came up behind us and gave a lengthy honk. The Waymo Driver, of course, did nothing but I was a bit unnerved. At the next stoplight, the driver of the truck pulled up next to us in the left turn lane and rolled down his window. Looking straight at me (I was in the front passenger seat), he yelled: “Hey d**kh**d, stay out of the left lane!” I responded by holding up my hands and saying: “I’m not driving.” That should have been obvious, of course, because to see me he had to look across the empty driver's seat, but it didn’t seem to lessen his anger and when the signal turned green he roared off.
I’m guessing this isn’t all that unusual. After all, similar things happen on a regular basis to human drivers who drive at or below the speed limit. Road rage is a serious issue and I suspect that robotaxis – and eventually other computer-driven vehicles – will only make the situation worse. Some drivers may go beyond honking and drive in an erratic manner knowing that the robotaxi’s computer will react quickly enough that they are unlikely to get in an accident. This seems like a recipe for disaster and I’m not sure exactly what can be done about it. Fortunately, robotaxis are decked out with cameras and other sensors so incidents of this type can be documented. Perhaps cities can use that documentation to issue traffic citations.
The Bottom Line
The key to public acceptance is going to be safety and robotaxis are off to a good start in that respect, although there have been some notable gaffes. A group of five Cruise robotaxis all stopped working at the same place last summer for example, blocking traffic in San Francisco for several hours. Just recently, another Cruise robotaxi drove at low speed into the back of a San Francisco bus. While these incidents (and others) sometimes seem inexplicably stupid, don’t forget that human drivers do inexplicably stupid things all the time. From the partial data that has been published so far, robotaxis seem to be involved in a relatively low number of accidents per mile driven, and those accidents tend to be relatively minor in most cases. The standard, in my opinion, is not that robotaxis should be perfect but that they should be as safe or safer than human cabbies and Uber drivers.
Waymo has been the most transparent about their driving record, perhaps because they seem to be doing the best job. Here are some of their recently published statistics for their autonomous robotaxi service:
In January 2023, Waymo passed 1 million rider-only miles (on top of millions of additional miles with a Waymo specialist in the car as a precaution).
There have been no injuries reported.
There have been 18 minor contact incidents and only two incidents which met the reporting criteria for the National Highway Transportation Safety Administration’s Crash Investigation Sampling System.
Fifty-five percent of all incidents were the result of a human-driven car colliding with a stationary Waymo vehicle.
There were no intersection-related incidents and no incidents involving pedestrians. 
It is important to point out, however, that these statistics are shaped by the fact that Waymo has been very cautious about the situations in which its vehicles are allowed to operate. It avoids many difficult situations, most hazardous weather environments, and does not drive outside of well-mapped areas where the potential hazards are understood. Still, the results are pretty impressive and seem to warrant expansion of robotaxi service to more locations.
Do not, however, expect robotaxis to be showing up in your city any time soon unless you live in California, Arizona, Nevada or Texas. Those are the only states that have been receptive to robotaxi pilot programs to date and even there expansion is piecemeal. In most states, the regulatory and legal framework for self-driving vehicles is virtually nonexistent. In addition, robotaxi service is likely to spread first into warm weather states that do not routinely experience snow and ice storms. Bad weather testing is taking place but those areas are likely to be low on the priority list for expansion. Finally, when expansion does take place to new cities, expect it to be limited to just part of the metro area and to be priced similar to Uber or Lyft. For the next several years, robotaxi service will still be in a highly monitored and carefully restricted test mode which means that cost reductions will not show up and the ability to drive to any and every address will not exist.
Despite the cautious approach that robotaxi firms are taking to expansion, it is working and it is expanding. And contrary to the claims of Elon Musk, robotaxi firms are miles ahead (excuse the pun) of traditional automakers. So while the robotaxi revolution might be delayed, it is most certainly not dead. It may be expanding at a glacial pace for now, but at some point that will change and cities need to be preparing for that eventuality.
I expect slow expansion in warm weather states to be the norm for the next 2 to 4 years. By that time, robotaxi firms will be rolling out purpose-built vehicles, safety concerns will be largely forgotten, and substantial progress will have been made on the pick up/drop off issue. At that point, other states will start enacting the necessary regulatory changes and robotaxis will start appearing in many more cities. The human support network that robotaxis need, however, means that they will not appear everywhere and will not cover the entire metro area even where they do operate. Profitability concerns will limit robotaxi service to areas where demand is high enough to ensure high utilization rates and efficient support services. The fear that some have expressed that empty robotaxis will be roaming around cities causing congestion is misplaced because empty vehicles will not be economically sustainable.
By 2030, robotaxis will be a useful transportation option for many people and will be considered a normal part of urban life. Continued innovation in the interface between the computer “driver” and human passengers (e.g. facial recognition or mobile video calls) will produce new types of service that don’t exist today. For example, parents might schedule a robotaxi to pick up their kids from school, including a quick video chat with Mom about “how was your day.” People who are elderly or disabled will have more mobility than ever, and tourists might ask their robotaxi for sight-seeing advice. In short, the combination of a self-driving vehicle and an AI powered computer presence will do more than your Uber driver ever could.
Aaron Gordon; “All the Big Claims Elon Musk Made About Tesla’s Autonomous Driving Plans”; Jalopnik; April 2019; https://jalopnik.com/all-the-big-claims-elon-musk-made-about-teslas-autonomo-1834238028
Adam Ismail; “Let’s All Laugh At These Bad Autonomous Car Predictions From Just A Few Years Ago”; Jalopnik; April 2021; https://jalopnik.com/let-s-all-laugh-at-these-bad-autonomous-car-predictions-1846690460
Harry Campbell; “Uber Statistics 2023: Drivers, Riders, Revenue & More”; The Rideshare Guy; March 2023; https://therideshareguy.com/uber-statistics/
“First Million Rider-Only Miles: How the Waymo Driver is Improving Road Safety”; The Waymo Blog; February 2023; https://blog.waymo.com/2023/02/first-million-rider-only-miles-how.html