Speed vs Safety: At What Speed Do AV Crashes Kill?

When we talk about autonomous vehicle crashes, we tend to focus on the total count: 6,215 incidents in the NHTSA SGO database. But that number treats a 5 mph parking lot bump the same as a 65 mph highway collision. The reality is that speed transforms what a crash is — from inconvenience to catastrophe. This analysis breaks down every reported incident by speed bracket to understand where AV systems are failing and what those failures actually mean.

The Low-Speed Majority

2,545 of 6,215 incidents — 41% — occurred at speeds under 25 mph. These are overwhelmingly minor collisions: parking lot scrapes, low-speed rear-endings in traffic, and gentle contacts during lane changes. Many involve Waymo and Cruise robotaxis operating in dense urban environments where traffic rarely exceeds 25 mph.

At these speeds, the physics of injury simply aren't there in most cases. The 0-10 mph bracket shows an injury rate of just 7.2%, and zero fatalities across 1,243 incidents. These crashes matter — they're still property damage, still scary, still failures — but they're categorically different from what happens at highway speed.

The Fatality Cliff

Something changes dramatically above 45 mph. The fatality rate, which hovers near zero below 25 mph, begins climbing steeply. At 46-55 mph, the rate reaches 2.46% — meaning roughly 1 in 40 crashes at this speed kills someone. At 56-65 mph, it peaks at 3.86%, with 24 fatalities from 621 incidents.

This is consistent with well-established crash physics. The kinetic energy of a vehicle scales with the square of velocity. A car traveling at 60 mph carries 4 times the energy of one at 30 mph. Human bodies simply cannot absorb the forces generated in high-speed impacts, regardless of whether the car has airbags, crumple zones, or any other safety feature.

Why This Matters for AV Safety Claims

The speed distribution of AV crashes has profound implications for how we evaluate safety. When a company reports its crash rate, the number is dominated by low-speed incidents that rarely cause injury. This mathematically dilutes the severity picture. A system with 1,000 parking lot bumps and 10 fatal highway crashes has a "low" overall fatality rate — but the highway performance is what actually matters for human survival.

This is particularly relevant for comparing ADS (fully autonomous) and ADAS (driver-assist) systems. ADS vehicles like Waymo operate almost exclusively at urban speeds under 35 mph. ADAS systems like Tesla Autopilot operate heavily at highway speeds of 55-75 mph. Comparing their raw crash rates without accounting for speed is comparing different risk universes.

The Injury Gradient

While fatality data understandably captures the most attention, the injury rate gradient is equally telling. At 0-10 mph, 7.2% of crashes cause injury. This climbs steadily to 42.6% at 46-55 mph and reaches 60.8% above 76 mph. In the highest speed bracket, more than half of all crashes produce injuries.

These injuries range from whiplash and bruising to spinal cord damage, traumatic brain injury, and limb loss. The higher the speed, the more likely a crash results in permanent, life-altering harm. When evaluating whether AV systems are "safe enough" for deployment, injury rates by speed bracket provide a much more honest picture than aggregate counts.

The Takeaway

Speed is the single most important variable in crash severity. AV safety reporting that doesn't stratify by speed is inherently misleading. A system that performs well at 15 mph in San Francisco traffic deserves credit — but that performance tells us nothing about its ability to prevent fatal crashes at 65 mph on I-95. Until manufacturers report crash rates by speed bracket, their aggregate numbers will continue to obscure the risks that matter most.