Waymo’s driverless future still has a very human backstop. When its robotaxis freeze in chaotic situations, police officers and firefighters are increasingly the ones who hop in, shift to manual override, and clear the scene so everyone else can do their jobs.
Consider a recent wildfire response near Redwood City, where California Highway Patrol (CHP) briefly reversed traffic on I-280 to evacuate drivers. A Waymo attempted to edge along the shoulder, got confused by the wrong-way flow, and stopped. After remote support failed to free the vehicle, a 911 call led a CHP officer to climb in and drive it out of the danger zone.
Why First Responders End Up At The Wheel
Emergency scenes break the rules that autonomous systems rely on. Wrong-way traffic, dark intersections during power outages, or police tape rerouting flows can invalidate carefully mapped assumptions. Waymo’s remote assistance can advise vehicles through oddities, but when seconds matter and lanes must be cleared, immobilized AVs turn into obstacles until a human moves them.
That’s played out across multiple cities in recent months. In Austin, an officer repositioned a Waymo to let an ambulance through during a mass-casualty response. In Atlanta, first responders disengaged a vehicle that turned into an active crime scene before a company team retrieved it. In Nashville, an officer manually steered a stuck robotaxi out of an intersection to restore traffic flow.
The Human Safety Net Behind Autonomous Rides
Waymo operates at considerable scale, reporting more than 400,000 paid rides each week across a fleet of roughly 3,000 vehicles. Edge cases are rare on a per-trip basis, but at this volume even a sliver of failures yields frequent real-world interventions.
Behind the scenes, a small cadre of “remote assistance” specialists stands by to help AVs reason through tricky scenarios. In a letter to Congress, Waymo said about 70 people monitor the fleet at any time, split between U.S. operations centers and teams in the Philippines. The company emphasizes that these workers do not drive the cars; they provide guidance, with claimed one-way connection latency of about 150 milliseconds in the U.S. and 250 milliseconds abroad.
When things escalate, an “event response team” coordinates with 911, manages post-collision protocols, and, if needed, dispatches on-the-ground help. Waymo also employs a roadside assistance operation—some personnel are contracted via firms like Transdev—that physically moves or tows vehicles. The company says it has built a straightforward process that lets first responders take control within seconds when imminent hazards demand it.
Even with human supervision, errors happen. The National Transportation Safety Board disclosed an Austin case where a remote assistant misread a stopped school bus situation, allowing a vehicle to proceed despite deployed lights and stop arms. The episode underscores a core tension: humans-in-the-loop can reduce risk but can also introduce new points of failure.
Training, Access, and Accountability for AV Incidents
Waymo says it has trained more than 30,000 first responders worldwide on how to interact with its cars and has collaborated with agencies on the manual override workflow. Company representatives stress they never ask passengers to take control and that they dispatch their own teams or tow partners when a vehicle needs to be moved.
City officials, however, are pressing for more. At a public hearing in San Francisco following a widespread blackout that stranded multiple AVs, emergency management leaders argued that police and fire crews are increasingly acting as de facto roadside assistance. Their message was blunt: public safety resources are finite, and AV operators need to shoulder more on-scene responsibility.
Officials want clear commitments—staffing targets, guaranteed arrival times for company responders, and fewer handoffs to 911. Regulators including the California Public Utilities Commission and city transportation agencies are also watching how robotaxis interact with emergency services as they set conditions for expanded service areas and hours.
What Would Reduce 911 Interventions for Robotaxis
Several levers could help. Technically, AVs need better detection of flares, tape, and handheld stop signs; geofenced slow zones during major incidents; and direct vehicle-to-everything links with patrol cars and fire rigs broadcasting closures in real time. Operationally, staging rapid-response crews near known hotspots and publishing service-level goals for clearance times could take pressure off first responders.
Policy can nudge this along. Cities and state regulators could require shared incident metrics, mandate regular joint drills with police and fire, and standardize the first-responder interface across AV fleets so any officer can quickly and safely move any robotaxi. Some companies are even exploring limited-use partnerships with gig workers for simple tasks—today that might be securing a door; tomorrow it could be repositioning a car a few feet clear of a hydrant.
Until those improvements mature, a paradox remains. The more conservatively AVs behave in unmodeled chaos, the more likely they are to stop and wait. And when a city needs the lane now, the fastest way to keep the public safe may still be the oldest one on the road: a trained officer sliding behind the wheel and driving.