Three thoughts on Waymo's recent traffic jams during a power outage in San Francisco.
First, these jams – in which Waymo's vehicles appeared to come to a stop in intersections and active travel lanes – show exactly why:
1) I have been pushing for four years to add a floor to the concept of minimal risk condition (MRC) in SAE J3016. (Unfortunately, this term may soon change to mitigated risk condition.) As I wrote earlier this year, "I do not know what Waymo’s automated driving system does while awaiting the provision of remote assistance. If the system continues moving in a safe manner or achieves a minimal risk condition, then that system would be properly classified as level 4. If, however, the vehicle would simply pause in an active lane of traffic in a way that a human driver would not, then the more appropriate classification would arguably be level 3."
2) The US Department of Transportation's now-disbanded Transforming Transportation Advisory Committee (TTAC) recommended, in late 2024, "scenario planning, break-the-glass plans for emergencies and other contingencies," and "analysis of the resilience of relevant transport systems, including the implications of potential skill or labor loss for evacuations and other emergency situations."
3) Matt Wansley and I called for holistic emergency planning that accounts for drastic changes in actual driving environments, loss of communications, overwhelmed remote assistants and retrieval crews, AV-related roadway obstructions, and mass dependence on AVs. This planning needs to involve developers as well as regulators, especially the local officials who are often some of the best informed and yet least empowered actors in this field. (First responders in cities such as San Francisco and Austin are effectively subsidizing both the development and the operation of automated vehicles. This valuable public contribution deserves more recognition.)
Second, to speculate:
If all wireless communications were down because of the local power outage, then Waymo's robotaxis would have been entirely unable to receive the remote assistance they seem to need to deal with these unusual – but still foreseeable – roadway conditions. (I have not asked Waymo whether it has redundant communications through multiple cellular networks as well as through alternatives to cellular service.)
If Waymo's remote assistants were still able to communicate with Waymo's AVs, then it's possible that nothing the remote assistants could recommend was deemed acceptable by the AVs themselves. (This tricky "I'm sorry, Dave" question merits more attention than it has received in the context of automated driving.)
Alternatively or additionally, the explanation might lie in an issue I've observed in the broader automated driving industry: inadequate systems for managing (i.e., assessing, triaging, queuing, assigning, and reassigning) requests for remote assistance that exceed the capacity for remote assistance. A suboptimal approach could create spiraling delays even greater than what one might expect from insufficient capacity alone.
Striking in Saturday's images is how some robotaxis were blocking other robotaxis. (I have also experienced this.) Mitigating the resulting gridlock could require unblocking particular AVs before or in conjunction with others. This may be difficult under an approach to automated driving (including remote assistance) that treats each AV as independent rather than as part of a broader system.
[An update from Waymo's December 23rd blog post: "While the Waymo Driver is designed to handle dark traffic signals as four-way stops, it may occasionally request a confirmation check to ensure it makes the safest choice. While we successfully traversed more than 7,000 dark signals on Saturday, the outage created a concentrated spike in these requests. This created a backlog that, in some cases, led to response delays contributing to congestion on already-overwhelmed streets."]
Third, Waymo should improve its communications in an entirely different sense. In addressing Saturday's apparent traffic jams – as well as recent incidents in which Waymo's robotaxis struck and killed two pets, failed to stop for school buses, and failed to detect a stowaway in the trunk, among others – the company should be much more forthcoming. It should explain what specifically happened and why (including what it does not yet know), what steps it will take to understand and reduce the risk of broadly similar and otherwise analogous incidents in the future, and when and how it has actually implemented those steps.
In Waymo's case, this approach should involve giving some of the company's engineers and researchers a much more significant role in engaging with local communities and with the public at large – both by sharing and by listening. Waymo's impressive academic work deserves more recognition and discussion, and Waymo itself is in a good position to bridge this gap between its technical research and its public engagement generally.