Emergency Vehicle Crashes: Tesla's Blind Spot

Emergency vehicles present a unique and dangerous challenge for autonomous and driver-assist systems. When fire trucks block lanes at accident scenes, when police cars sit on highway shoulders with lights blazing, when ambulances are parked at emergencies — these are precisely the situations where automated driving systems need to respond correctly. Our analysis of NHTSA's SGO database found 369 crash narratives that mention emergency vehicles, police, fire trucks, or ambulances.

The Scale of the Problem

369 incidents is not a small number. It represents roughly 6% of all AV/ADAS crashes in the database. These crashes produced 7 fatalities and 319 injuries— an 86% injury rate that far exceeds the overall database average. Emergency vehicle scenes involve complex, unusual road configurations that stress-test every aspect of an automated system's perception and decision-making.

Who's Crashing?

The manufacturer breakdown reveals an important nuance. Zoox leads with 111 incidents, followed by Cruise (57), Waymo (48), and General Motors (33). Tesla accounts for 17 incidents. But raw counts can be misleading — Zoox, Cruise, and Waymo operate urban robotaxi fleets that frequently encounter emergency scenes in dense city environments. Many of these are low-speed encounters where the AV appropriately yielded but was struck by another vehicle.

Tesla's 17 incidents, while fewer in number, represent a qualitatively different failure mode. Tesla crashes disproportionately involve high-speed highway impacts with stationary emergency vehicles — the system fails to detect or react to a massive, brightly-lit obstacle directly in its path. NHTSA opened formal investigation PE 21-020 specifically targeting this pattern.

The Stationary Object Problem

The technical challenge is known as the stationary object detection problem. Historically, radar-based systems filtered out stationary objects to avoid false positives from bridges, signs, and overpasses. When a vehicle approaches a stopped fire truck at 65 mph, the system must distinguish it from background clutter — and it must do so with enough time to brake from highway speed.

Camera-based systems face their own challenges: a stationary emergency vehicle can appear suddenly as the car rounds a curve or crests a hill. The visual signature of flashing emergency lights, which make vehicles unmissable to humans, may paradoxically confuse computer vision systems expecting steady-state visual patterns.

Two Types of Emergency Vehicle Crashes

The data reveals two distinct crash patterns. The first is the urban encounter: a robotaxi operating at low speed encounters an emergency scene and either stops abruptly (getting rear-ended) or attempts to navigate around it (sideswiping another vehicle). These are generally low-severity and represent the operational reality of driving in cities with active emergency responses.

The second pattern is the highway strike: a vehicle on ADAS (typically Autopilot) traveling at highway speed fails to detect a stationary emergency vehicle and crashes at full speed. These are the catastrophic incidents — the ones that kill people — and they represent a fundamental perception failure that no amount of driver monitoring can fully mitigate when reaction time is measured in fractions of a second.

The First Responder Perspective

First responder organizations have raised alarms. The International Association of Fire Chiefs has flagged the risk, and some departments have modified blocking protocols to account for AV limitations. This represents a remarkable inversion: humans adapting their safety procedures to protect themselves from the limitations of machines that were supposed to make roads safer.

What Needs to Change

Emergency vehicle detection should be a solved problem. These are the largest, most visible, most distinctively-marked vehicles on the road. Any autonomous system that cannot reliably detect and respond to them has no business operating at highway speed. The 369 incidents in this database — and the 7 deaths they've caused — are a clear signal that the industry has not yet met this basic safety threshold.