(Image: Waymo)
By Colin Barnden
What’s at stake:
First Cruise, now Waymo. When will AV developers learn that publishing statistics to make their machine driver look safer than human drivers both insults our intelligence and does nothing to make roads safer.
At the beginning of September, Waymo released updates to its Safety Impact assessment with new statistical data and conclusions. The findings, shown below, are incredible.
So safe is the Waymo Driver that for over 22 million cumulative “rider-only” (driverless) miles traveled, airbag deployment crashes were 84 percent lower than for comparable human-driven miles, and 73 percent lower for “injury-causing crashes.”
The key takeaway: The Waymo Driver is already safer than an average human driver.
Not so fast. By following the GM Cruise playbook, Waymo is resorting to questionable methods to achieve its messaging goals.
First, let’s put that figure of 22 million rider-only miles into context using data from the Federal Highway Administration which reveal annual vehicle miles traveled on U.S. roads totaled about 3.2 trillion in 2022. Yes, that’s trillion, or a million million.
Waymo’s selective comparisons of human driving with the Waymo Driver is not independent analysis and must be treated with caution.
(Editor’s note: Phil Koopman, prof. of Carnegie Mellon Univ., just publised his detailed analysis on “The New Way Safety Dashboard” in his substack newsletter)
Second, if the Waymo Driver is already so incredibly safe, why are Waymo’s robotaxis making the news for driving headlong into oncoming traffic?
Think that is a one-off? Here’s another incident.
In recent months, San Franscisco resident Sophia Tung has livestreamed a webcam feed from her apartment overlooking a Waymo robotaxi depot.
Let’s watch an example of The World’s Most Experienced Driver™ in action.
How can Waymo’s statistics make the Waymo Driver appear safer than a human driver while visual evidence confirms it is quite unbelievably stupid? The answer is: It depends on the aspect of the driving task being assessed.
Rasmussen’s SRK framework
Jens Rasmussen was a system safety, human factors and cognitive systems engineering researcher who was highly influential within the fields of safety science, human error and accident research. One key area of his work was the skills, rules, knowledge (SRK) framework.
The task of driving can be broken down and analyzed using these four distinct competencies covering skills, rules, knowledge, and expertise.
Missy Cummings, a professor at George Mason University and previously a safety advisor at NHTSA (National Highway Traffic Safety Administration), could be credited with extending Rasmussen’s SRK framework to include a fourth category, covering expertise.
Therefore, the task of driving can be broken down and analyzed using these four distinct competencies covering skills, rules, knowledge, and expertise.
Let’s very briefly review each one.
- Skills: This can be seen as the most basic tasks of driving, such as maintaining a safe distance from other vehicles and road users, and of keeping a vehicle centered in its lane.
Automation is vastly more reliable than humans at extremely boring and repetitive functions such as managing distance control and lane centering, provided driving conditions remain entirely benign. Automation can also respond much faster than a human driver in the event of an imminent collision.
- Rules: Examples of rule following include driving on the correct side of the road; observing speed limits; stopping at red lights and stop signs; following one-way signage; obeying “no entry” or “no turn” signs; understanding “no right on red” signs; navigating four-way stops; navigating roundabouts and; yielding to pedestrians.
Many human drivers, particularly young males, willfully ignore driving rules such as speed limits and red lights. However, both Tesla’s Autopilot and FSD continue to demonstrate failure to follow basic rules of the road; so too Waymo robotaxis ignoring “no left turn” signs and bus lane rules, as shown below.
- Knowledge: Examples include correctly interpreting an “80” speed sign in a residential street as kids having fun; and not driving into utility poles.
Competent human drivers use knowledge-based reasoning to demonstrate judgment under uncertainty. A dude standing on a sidewalk wearing a t-shirt printed with a stop sign is a dude standing on a sidewalk wearing a t-shirt, not a stop sign.
Let’s revisit The World’s Most Experienced Driver™ to assess its competency at knowledge-based reasoning, using this video.
- Expertise: This is best described as first-principles reasoning, or the ability to consider every aspect of a novel problem and to improvise a suitable solution. A perfect example is provided from the world of aviation, where Capt. Chesley “Sully” Sullenberger improvised landing on the Hudson River following double engine failure resulting from a bird strike.
Cummings’ SRKE model therefore shows us that automation vastly outperforms human ability for precisely definable skills such as longitudinal distance and lateral lane control. But as uncertainty increases through rule following towards the knowledge and expertise necessary to perform the driving task in everyday conditions, so human drivers vastly outperform machine drivers.
Waymo’s latest Safety Impact assessment therefore only tells us the Waymo Driver is better than an average human driver at reacting to an imminent collision. That is what is shown in its headline findings, with fewer airbag deployment crashes and fewer injury-causing crashes reported.
Stacked with many tens of thousands of dollars of whopping-TOPS compute processors and state-of-the-art multi-modal sensor suites on every vehicle, Waymo’s robotaxi fleet frankly should be able to react sooner and stop faster than an average human, with average reaction times, driving an average twelve-year-old car or truck, with little or no driver assist.
The critical safety advance therefore won’t come from the Waymo Driver, but mass-market deployment of technology to sense an imminent collision, compute a safe response, and act to apply maximum braking force significantly faster than any unaided human driver can.
That technology is not unique to Waymo. We already know it by the name automatic emergency braking, or AEB.
Making human drivers safer
The AV industry missed its window of opportunity to dominate the road safety argument, and the regulatory environment has already moved on. NHTSA issued FMVSS #127 for AEB in April, which specifies performance minimums and requires vehicle manufacturers to make AEB standard in all new cars and light trucks starting in 2029.
Europe has gone further and mandated AEB for all vehicles with four or more wheels in the revised General Safety Regulations (GSR), which also includes trucks, buses and coaches, in addition to passenger vehicles.
AV suppliers repeatedly state that machine drivers are safer than humans because machines do not drive drowsy, distracted, or drunk. But here too safety legislation is already far ahead of the tired gaslighting from the AV industry.
Mitigating human distraction and drowsiness is the primary role of a vision-based driver monitoring system (DMS), and Euro NCAP began awarding points for DMS in its 5-star safety rating system last year. It could add cognitive distraction and impairment detection to its revised 2026 protocols, which are due to be published soon.
Starting July, Europe’s GSR also required distraction and drowsiness monitoring, and the mandatory requirements may be extended further to incorporate aspects of the 2026 Euro NCAP protocols as soon as 2027.
However, the most relevant policy advance came at the end of last year, when NHTSA issued an Advanced Notice of Proposed Rulemaking (ANPRM) for impaired driving. Language in the ANPRM references distraction, drowsiness, and drunk driving prevention and so is likely to incorporate many aspects of Europe’s GSR when it is published as an FMVSS, which is due before the end of this year.
Human reasoning remains vastly more advanced than machine reasoning in ambiguous, uncertain and extreme driving situations. The more uncertain the circumstances, the more that safe driving is reliant upon expertise and knowledge over rules and skills.
Having focused entirely on developing technology to replace humans as drivers, Waymo has no role to play in regulatory efforts to make human drivers safer. Like Cruise before it, Waymo’s campaign of denigrating human drivers reflects poor judgment, and does nothing to make public roads safer.
Bottom line:
Robotaxis will play a future role in urban transit systems, alongside other ride hail services and public transportation. However, the greatest progress towards saving lives on U.S. public roads will result from NHTSA mandates for DMS and AEB in human-driven vehicles.
Colin Barnden is principal analyst at Semicast Research. He can be reached at [email protected].
Copyright permission/reprint service of a full Ojo-Yoshida Report story is available for promotional use on your website, marketing materials and social media promotions. Please send us an email at [email protected] for details.