Featured Image

California Dreams of Autonomous Vehicles: Lessons from the summer for avoiding an AV winter

7 min read

The regulatory fight over expanded robotaxi deployment in California, and the implications for GM's Cruise, highlight what we as an industry need to take seriously.

Who would have thought that tech regulation could unfold like some will-they or won't-they reality television program. But it is California, the home of Hollywood, we are talking about here.

Specifically, this summer saw the slow-burn decision about whether to allow commercial robotaxis to deploy more widely on the streets of San Francisco and San Mateo counties.

Approval was initially taken for granted in tech-loving Silicon Valley. But then persons and organizations with complaints and concerns came out of the woodwork like it was the last episode of Seinfeld.

Their stories, and the story of Californian robotaxis, are important. California is a bellwether for autonomous vehicle (AV) deployment. Each detour, U-turn, and dead-end in their journey there can prepare us for our own autonomous future and help prevent an AV winter.

A seemingly endless summer

Every drama has a backstory. Ours begins in the spring of 2018, when the state-level California Public Utilities Commission (CPUC) authorized the first autonomous vehicles to provide free tests rides as part of its Autonomous Vehicle Passenger Service Programs. This led to vehicles being deployed: allowing testing and data collection for the AIs and trust-building with the community.

A couple of years later the CPUC expanded the programs to open up the possibility for companies to charge for rides. The first permits for commercial driverless AV service were issued in 2022, albeit with restrictions on operating times, speeds, safety drivers, passengers, and ability to monetize.

The next step in the CPUC's plan was to expand hourly operations and service area, remove limits on the number of robotaxis, and allow operators to charge for all rides. Cruise and Waymo, the leading operators in the state, would be the immediate beneficiaries. Accordingly, in anticipation, they began rapidly scaling their operations throughout 2023. It seemed the good times for AV firms would never end.

Seasons change

But the summer of AV expansion was not to be and long in advance of the 29 June vote the antagonists of our story made their voices heard. The interest in robotaxis, and amusement when they malfunctioned, turned to frustration as commuters blamed them, fairly or not, for traffic jams and public transit delays.

Citizens' concerns were echoed by city-level emergency services, who said the vehicles were interfering with their safety critical operations. Safety is a big argument for more AVs, so if their own safety is offset by making others unsafe, it would be a serious blow to their rationale. Yet even the robotaxis own safety record was being questioned as they were also involved in numerous road incidents.

How numerous and accurate were these accusations? Unfortunately, there is no way to be certain. San Francisco's transportation agency alleges the number of incidents is underreported. Even amongst those reported, there is little clarity on whether an “incident” was an accident and the AV's roles in them. The AV's interference with emergency vehicles and scenes is more quantified, but the implication and severity are not, with the incidents characterized by conditionals: if, could have, what might have been.

With no clarity on AV's present, the clouds were already appearing on the horizon of AV's future.

Summer comes late

The aforementioned concerns became a rationale for twice delaying the hearing: allowing more time for companies to submit evidence of their record, concerned parties to lodge their protests (most notably San Francisco's transit agency and fire and police departments), and the committee to review what was received.

Decision day finally arrived on 10 August, with a hearing that saw six hours of testimony from hundreds of stakeholders. Their concerns generally fell into one of five categories:

  • Commuting specific concerns: These included complaints of AVs blocking roads, causing traffic jams, and interfering with public transportation.
  • Economic concerns: Including loss of jobs as well as a more general unwillingness to support 'big tech'.
  • Libertarian concerns: Worries that the data from AVs would support the surveillance state. 
  • Interest-group specific concerns: Notable here was the disabled community, which, while standing to benefit much from AVs, highlighted glaring oversights in vehicle accessibility.
  • Safety concerns: Especially, as noted above, those expressed by fire and police departments. These included impeding emergency vehicles and infringing on emergency scenes.

Against such a range of attacks, the future of CA AV looked grim. The last concern, safety, seemed the most salient. A statement by Commissioner Genevieve Shiroma showed how hard the case was for the AV firms I this regard: "All it takes is one real life example of a driverless autonomous vehicle that prevents a first responder from doing its job in real time that convinces me that we should not approve citywide deployment."

This bias toward edge cases, and even a sample size of only one, is an impossible criterion to overcome. Nonetheless, in the final vote of the four commissioners, the expansion was approved three to one.

The sky turns grey

But soon long-delayed summer turned to autumn, as a Cruise vehicle was involved in an October traffic incident. Details have been reported elsewhere; what matters here is that even though the robotaxi was not the cause, and followed its programming as to how to react to an incident, a person's injuries were exacerbated by the AV's actions and Cruise's operational permits were suspended by the state of California.

Yet the leaves had just begun to fall, and in the following days and weeks Cruise suspended all operations in the US, recalled hundreds of vehicles, halted production of vehicles, reported a loss in the hundreds of millions of dollars, and (most recently as of this writing) saw its founder and CEO step down.

Is winter coming?

While the AV companies and their supporters won the initial round, the struggle to get there and Cruise's quick relegation to the penalty box shows it is too early to take a victory lap.

The commission was quite explicit that it expects the companies to address the concerns of the emergency services and law enforcement. Subtle hints were given that this decision is just as easily revoked as it was given. And indeed, they showed with Cruise that this was the case.

Then there is the matter of how this de jure decision plays out in de facto reality. The CPUC may have authority based on the bureaucratic hierarchy, but the reality on the streets is often controlled by authorities much closer to them. There is a truism that revolutions fail if the municipal governments oppose them.

The local government in our story was not won over by the AV firms. It even has a resolution approved (but as of this writing unsigned) waiting in the wings to slap them with so many restrictions as to make doing business unfeasible. There are a million ways for a city to spoil the state's and AV companies’ plans. 

And now perhaps another level may get involved, one superseding the CPUC. In November dozens of unions signed a letter calling for federal regulations of AVs, which they called “unsafe and untenable”. Given the signatories of this statement though, a strong economic rationale unpins their concerns alongside the safety case.


Many lessons could be drawn from this summer's events. We choose here to focus on three. These are ones we see as the most serious, but also the most easily resolved through cooperation and compromise amongst defined stakeholders. Larger issues encompassing all of society, such as economic and libertarian concerns, will require a much longer conversation.

  • Lesson 1: Transparency is required from both sides. The firms need to do better sharing their data and reporting incidents, as often called for by city officials in our case here. But government officials need to do better on classifying incidents and assuring it will use shared data for ex ante policy making, city planning, and CV regulation and not only for ex post remedial legal action. The fear of litigation is strong in the US and without clarity and legal certainty, actions obviously in everyone's best interests are often not taken. 
  • Lesson 2: Companies need to worry less about some inchoate general level of public acceptance and trust in AV technology and focus on socially and culturally salient sub-groups: prove to them your tech is worth adopting and enlist them as allies. In the US, first responders fall into this group. The logic of "protect the protectors” is sacrosanct.  Historically disadvantaged or marginalized persons are another important group. If your company re-creates the past, it has no future.  
  • Lesson 3: Probably most importantly, firms need to improve their safety argumentation. This is not easy: without vehicles deployed and real-world accident rates they can't rely on actuarial sciences/statistics. But having no definition of what is "safe”, no bars, has allowed opponents of AV technology to turn the companies’ safety claims against them: one failure is enough. Without data (see Lesson 1) allowing statistical evidence, the anecdotal exemplars prevail.

Integrating rapid development in the field of AI-driven autonomy into historically more slow-moving vehicle development may require a rethink of how companies deploy the resulting combination – autonomous vehicles. Reconceptualization may be required of data as a proprietary or public good, of how mobility solutions should be marketed and deployed, and of how safety can be proved and improved simultaneous to testing and (re-)training in deployment.

Perhaps winter can't be avoided, but with the right foresight, just like with well-trained autonomous vehicles, it can be safely navigated until spring returns.

© neurocat GmbH

Back to top Arrow