Waymo: Service resumed?
During the weekend of December 20, 2025, San Francisco experienced a widespread power outage that paralyzed large parts of the city. Amidst the chaos of dark intersections and honking cars, Waymo's robotaxis, which came to a standstill in many places, were particularly noticeable. What at first glance appeared to be a failure of autonomous technology turned out to be a stress test for the interplay of autonomous driving and urban infrastructure. This incident highlighted the dependencies of autonomous systems on external conditions and raised questions about safety and regulation in exceptional situations.
Source: Own visualization
Introduction
The power outage in San Francisco on December 20, 2025, during which dark intersections and honking cars characterized the cityscape, , posed a special challenge for the robotaxis operating there. Waymo vehicles stopped in many places, reigniting the discussion about autonomous driving and its limits. This incident, which seemed like an "edge case," was actually a stress test for an entire mobility concept , where autonomous technology meets suddenly unreliable infrastructure., The power outage in San Francisco
The power outage in San Francisco
On Saturday, December 20, 2025, a major power outage hit large parts of San Francisco , temporarily paralyzing about a third of PG&E's customers in the city. At its peak, PG&E and the city spoke of around 130,000 affected customers; by Sunday morning, according to PG&E, about 110,000 were already reconnected, while tens of thousands remained without power for a time. The outage led to disruptions in city life, including traffic jams and sometimes closed businesses. . The problem became particularly visible at intersections because traffic lights failed or remained dark.
The cause of the blackout was a fire in a substation, which, according to PG&E, caused "significant and extensive" damage. . Several reports located the event at the substation near 8th Street and Mission Street, where firefighters were deployed due to a fire. The San Francisco Chronicle also described a (co-)triggering fire at a PG&E substation in this area, while PG&E continued to investigate the overall cause.
Waymo's response and behavior
Waymo didn't come into play as a trigger, but as an amplifier in the street scene. Videos and photos showed Waymo vehicles stopped at dark intersections, , some with hazard lights on, while other cars had to navigate around them. The Guardian reported on "clusters" of several immobile robotaxis standing at intersections in the rain, causing traffic jams. ABC7 also documented that Waymo vehicles were stuck at intersections, , after traffic lights failed.
Source: automobilwoche.de
Waymo vehicles are a familiar sight in the streets of San Francisco and other cities where the autonomous ride-hailing service operates.
Waymo then temporarily suspended its driverless ride-hailing operation in San Francisco on Saturday evening. A Waymo spokesperson spoke of a temporary suspension of services "due to the widespread power outage" and emphasized coordination with city authorities. According to the San Francisco Chronicle, Waymo proactively paused; most ongoing trips were completed before vehicles safely returned to the depot or stopped. TechCrunch wrote that Waymo only announced the resumption of service late Sunday afternoon (December 21, 2025). The San Francisco Chronicle also reported that Waymo resumed operations on Sunday.
Autonomous driving and infrastructure
When a robotaxi stops during a blackout, it's not necessarily a failure, but often intentional. When a traffic light fails, California law expects people to treat the intersection like an all-way stop situation: stop and only proceed when it's safe. Caltrans formulates the same basic rule for power outages: Flashing red or completely dark signals are treated as stop signs. . According to the Chronicle, Waymo says the vehicles are designed precisely for this: non-functioning signals are treated as four-way stops.
Source: latimes.com
The complex environment of a large city presents autonomous vehicles with constant challenges that require precise responses.
However, a single dark intersection is different from a city district full of failed traffic lights – plus rain, plus holiday traffic, plus people improvising simultaneously. In such a situation, "cautious" can quickly become "too cautious": if the system does not recognize a clear right-of-way situation, stopping is often the safest option – but the worst for traffic flow. People often resolve dark intersections through eye contact, gestures, a quick wave of the hand, or "bold" rolling forward – things that are not always legally clear, but practically effective. A robotaxi can only afford such implicit deals if it reliably recognizes and correctly assesses them – and that is precisely what is extremely difficult in unclear situations. San Francisco Mayor Daniel Lurie, according to the Guardian, asked the public to avoid driving if possible due to failed traffic lights; at the same time, police officers were dispatched to major intersections. Reuters also reported that official bodies advised avoiding unnecessary trips and treating failed signals as four-way stops. . If the city itself is deploying people at intersections, it shows: without functioning signal infrastructure, traffic quickly becomes manual crisis management – and autonomous systems are then just part of the problem, but not automatically the solution.
Technology and dependencies
Waymo relies on a sensor mix of lidar, cameras, and radar, describing it as the basis for capturing the environment in three dimensions and navigating reliably. Even in Waymo's help section, lidar, cameras, and radar are listed as core sensors for the "Sense" module. Nevertheless, daily operation depends on more than just "seeing." In a recent technical contribution, Waymo describes that systems, among other things, generate or use high-resolution maps/map representations, , to predict behavior and plan trajectories. When traffic lights fail, not only is a visual signal missing, but often the entire "order layer" of an intersection: who has right-of-way, who goes first, who is seen, who hesitates?
Source: vuink.com
The advanced sensor technology on the roof of Waymo vehicles is the core of autonomous driving technology and crucial for navigation and safety.
The Guardian puts it succinctly: the outage made it visible how much autonomous vehicles depend on surrounding infrastructure – things that Waymo does not control. Also interesting is how autonomy is "backed up" in practice. Waymo describes its remote assistance program ("Fleet Response") as a way to receive additional contextual information from human agents, without them controlling the vehicle. This is not "remote driving," but a safety net for edge cases – and a city-wide blackout produces plenty of such edge cases. receive additional contextual information from human agents, , without them controlling the vehicle. This is not "remote driving," but a safety net for edge cases – and a city-wide blackout produces plenty of such edge cases.
Regulation and political implications
Waymo does not operate in San Francisco in "test mode," but as a commercial service. In 2023, the California regulatory authority CPUC already approved Waymo to charge fares for passenger transport. with driverless vehicles in San Francisco. In parallel, the California DMV publishes ODD maps (Operational Design Domain) approved for Waymo and framework conditions for driverless testing and deployment. And Waymo openly advertises the ride-hailing option in San Francisco as an available, autonomous service.
In this context, a blackout standstill has a different effect than a common "bug." As soon as a system becomes part of public essential services – mobility, emergency routes, traffic safety – every disruption becomes a question of duties: who must guarantee what, and what happens when the city itself (power/traffic lights) fails? The San Francisco Chronicle explicitly states that the blackout is reigniting the debate about how autonomous vehicles function during large-scale infrastructure failures.
The power outage in San Francisco was not a "Waymo problem," but it made a real autonomy problem visible: no matter how well robotaxis can drive – if the city around them loses its basic signals, routine traffic becomes an exceptional situation. Waymo responded with a pause and only resumed service when the situation stabilized – a sensible step that also shows how closely autonomy and infrastructure are intertwined in everyday life. The lesson is uncomfortable but clear: not only the car must be autonomous – the city must also become more resilient to failures if we want to truly treat autonomous mobility as a reliable part of the system.