Data Deep Dive

Where Do AV Crashes Happen? A Roadway Analysis

We analyzed all 6,215 reported AV and ADAS incidents to understand which roadway types see the most crashes โ€” and what the numbers reveal about automated driving risk.

1,907

Street Incidents

Most common roadway type

1,804

Highway / Freeway

Nearly equal to streets

1,204

Intersection Crashes

19% of all incidents

203

Parking Lot Incidents

Low speed, still dangerous

Where a crash happens tells us a lot about how it happens. NHTSA's SGO data categorizes each incident by roadway type, giving us a clearer picture of the environments where autonomous and driver-assist systems struggle most.

Streets vs. Highways: An Almost Even Split

With 1,907 incidents on streets and 1,804 on highways or freeways, the split is remarkably close. This challenges the assumption that most AV crashes are highway pileups. In reality, urban streets โ€” with their pedestrians, cyclists, traffic signals, and parked cars โ€” produce just as many (if not more) incidents.

For ADAS systems like Tesla Autopilot, which were originally designed as highway assistants, the high count of street-level crashes suggests widespread use outside intended operating domains. For ADS systems like Waymo, street incidents reflect the urban environments where robotaxis actually operate.

Intersections: 19% of All Crashes

1,204 incidents occurred at intersections โ€” roughly 19% of all AV/ADAS crashes. Intersections are notoriously challenging for automated systems: they require interpreting traffic lights, yielding to turning vehicles, reading pedestrian intent, and navigating complex right-of-way rules. This data confirms that intersections remain a major weak point.

Parking Lots: Low Speed, Real Risk

203 incidents in parking lots may seem small relative to the total, but these are low-speed environments where automation should arguably perform well. Smart Summon and self-parking features are specifically designed for these settings, yet they still produce crashes โ€” often involving pedestrians or other vehicles at very low speeds.

The Long Tail: Rural Roads, Traffic Circles, and More

Rural roads account for 104 incidents, traffic circles for 10, unpaved roads for 4, and parking garages just 1. These numbers are small but meaningful โ€” they show AV systems encountering conditions they may not be well-trained for. Rural roads, in particular, pose unique challenges: unmarked lanes, wildlife crossings, and variable road surfaces.

What This Means

The near-parity between street and highway crashes is a critical insight. It means safety efforts can't focus on one environment alone. Intersection incidents, at nearly 1 in 5 crashes, represent a clear engineering challenge that manufacturers must solve before wider deployment. And parking lot crashes โ€” though fewer โ€” undermine consumer confidence in the "easy" use cases that automakers market most heavily.

Incidents by Roadway Type

Roadway Distribution (excl. Unknown)

Speed Distribution Across All Roadways

Speed data helps contextualize roadway analysis โ€” low-speed parking lot crashes vs. high-speed highway incidents.