Why Robotaxis Have an Inherent 3x Advantage for Reducing Passenger Harm
BUT -- this has nothing to do with being a potentially better driver
If robotaxis were exactly as good as ride hail drivers, we would expect to see a only one third the occupant harm. It’s not a question of magic technology — it’s a simple matter of passenger ratios.
“It’s a simple question of weight ratios”— Monty Python and the Holy Grail
Let’s use some round numbers here to make the point: 1 passenger on a trip, and half of miles are deadhead miles (no passenger), assuming equal crash rates.
If a ride hail vehicle is empty half the time, on average it has 1.5 occupants. (The driver is always there + a passenger half the time.)
If a robotaxi vehicle is empty half the time, on average it has 0.5 occupants. (Empty half the time; a passenger half the time.)
So the expected injury rate per mile is 3x higher for the ride hail vehicle than the robotaxi — simply because there are 3 times as many people in a ride hail vehicle (1.5) than in a robotaxi on average (0.5).
So is about exposure, given equal crash rates, and not being a “better driver” somehow. To show this is not simply a word play trick, here are simple numbers based on those simplified assumptions:
Robotaxi runs 200 million miles at one fatal-severity crash per 100 million miles. On average one of the two expected fatal-severity crashes it is driving empty, so 1 fatality.
Ride hail runs 200 million miles at the same crash rate, also with two fatal-severity crashes. One crash will have only the driver, but the other crash will have passenger+driver, so 3 fatalities in all. 3x worse. (Yes, this is simplified to illustrate the point; might well be 1 fatality + 1 injury, etc.)
Looking at this another way in terms of # of crashes instead of # of people harmed:
If half the robotaxi miles are empty, half of their crashes won’t cause passenger harm — because there will be no passengers to get harmed. (Thus, half the injury crash events.)
HOWEVER — this does not mean we should flock to robotaxis as a compelling safety advantage based on this statistic alone, as much as the robotaxi companies want to sell us based on this type of number.
This number neglects pedestrian harm, harm to other vulnerable road users, and harm to people in other vehicles. It also neglects more indirect harms such as luring people away from public transit which is easily 10x safer than cars. And my 3x multiplier is based on round numbers — so real numbers will vary. I’m illustrating the principle rather than making a specific claim about a specific company who won’t reveal enough of their data to do the calculations for real.
A related claim is that robo-delivery vehicles are safer because crashes won’t injure the driver in a driverless cargo vehicle. But what about everyone else on the road if there are swarms of these things doing package delivery?
So if you want to support robotaxis, you might say that they potentially have a 3x safety advantage for occupants.
But if you care about net safety, when you see a robotaxi boasting 2x or 3x fewer passenger injuries the reaction is “I should hope so — because you’re running around empty so much” and “but what about injuries to other road users, and increased net injury due to pulling people away from mass transit? Where are those numbers?”
Or perhaps a company is only measuring miles with a rider (which avoids this pitfall), but is therefore excluding potential harm to other road users when they are deadheading. Also a problem.
And if you see a robotaxi company advertising better than a human driver — but only counting the at-fault crashes — then it sounds to me like their driving might be a lot worse than a ride hail driver.
I’ve seen combinations of all of the above. I expect them to continue. And the industry’s idea of transparency is barely beyond “as required by law” so it is infeasible for us to know how things are going ourselves beyond watching the headlines for crash stories.
The take-away: probably any single number you see advertised by a robotaxi company doesn’t mean what their marketing people want you to think it means. I’ve spent time digging through the reports, and in the end what I come away with is: (1) the claims in studies are so narrow and convoluted that they are difficult to align with any meaningful real-world outcome other than providing grist for a marketing claim, and (2) the industry is nowhere near being able to support the claims they are making about reducing fatalities any time soon — still a few hundred more million miles to go until we know how this safety experiment on public roads will turn out.
Source notes: Someone (and I wish I could remember who to credit them) mentioned in a social media comment that among the many flaws and weaknesses in robotaxi safety studies is not accounting for deadheading miles. Sounds like something Missy Cummings would say, but I’m not sure. And there was another study talking about robotaxis running around empty half the time as a congestion issue. Sounds like something David Zipper would say, but I’m not sure. But that was not today, although these two ideas converged this morning — so I’m unable to track them down in the ephemeral mess that is social media. If you want to claim credit please leave a link, even if it is multiple people! These topics are worth knowing more about.
One study found that 41% of ride hail miles were deadheading, so “half” is in the ballpark. Robotaxis tend to do a lot of on-road “testing” so that likely inflates the number for them.
And as to robotaxi claims of lower crash rates in general, you have to be very careful when they exclude lots of crashes for <reasons> and amplify human crash rates for their baseline for <other reasons>. Lots of games being played here — but that is a different posting topic.