Pedestrian Safety and Autonomous Vehicles: The Data We Have

Pedestrian safety is the most emotionally charged topic in autonomous vehicle safety. Every AV company promises to reduce the 7,500+ annual pedestrian deaths in America. But the incident data tells a more nuanced story — one where 68 total AV/ADAS fatalities include both occupants and vulnerable road users, and where the technology's performance with pedestrians remains a work in progress.

What the AV Data Shows

The NHTSA SGO database tracks crash severity (fatal, injury, no injury) but doesn't always specify victim type (occupant vs. pedestrian vs. cyclist). However, we know several high-profile pedestrian incidents:

• 2018 Uber ATG (Tempe, AZ): A self-driving Uber struck and killed a pedestrian crossing the street — the first fatal AV-pedestrian crash. This predates the SGO.
• Cruise (San Francisco): Multiple incidents involving pedestrians, including the 2023 dragging incident that led to Cruise's shutdown. 155 Cruise incidents, 0 fatalities.
• Waymo: With 1,729 incidents primarily in pedestrian-heavy San Francisco and Phoenix, Waymo's 2 fatalities include incidents where other road users were involved.
• Tesla: Several of the 56 Tesla fatalities involve pedestrian or cyclist strikes where Autopilot/FSD failed to detect or react to a vulnerable road user.

Urban vs. Highway

Pedestrian risk is concentrated in urban environments — exactly where ADS robotaxis operate. San Francisco's 1,170 incidents occur in a dense urban setting full of pedestrians, cyclists, and scooters. Phoenix's 261 incidents include the sprawling, pedestrian-unfriendly infrastructure that was the setting for the Uber ATG fatality.

Tesla's ADAS incidents are spread across all environments including highways (where pedestrian encounters are rare) and suburban areas (where they're more common). The Model 3's 29 fatalities and Model Y's 18 fatalities likely include a mix of occupant and non-occupant casualties.

The Technology Gap

Detecting pedestrians is one of the hardest problems in autonomous driving. Pedestrians are small, move unpredictably, wear varying clothing, and appear in complex backgrounds. Lidar excels at detecting pedestrians because it creates precise 3D point clouds regardless of lighting. Cameras struggle in low light, glare, and occlusion. Tesla's camera-only approach is at a theoretical disadvantage for pedestrian detection — which makes NHTSA's PE25012 investigation (camera visibility failures) particularly relevant.

The Promise vs. Reality

AV companies claim their technology will eventually eliminate the ~7,500 annual pedestrian deaths caused by human drivers. The data so far suggests fully autonomous ADS systems are relatively cautious around pedestrians (Waymo and Cruise have had very few serious pedestrian injuries given their urban exposure). But ADAS systems like Tesla's, operating in the messy real world with distracted human "supervisors," haven't yet proven their pedestrian safety credentials.