Petrobras is world’s largest operator of floating production platforms
Norwegian Cruise Line selects Rolls-Royce permanent magnet tunnel thruster

UMTRI researchers suggests safety expectations for self-driving vehicles may be overblown

In a new white paper, a team from the University of Michigan Transportation Research Institute (UMTRI) discussed issues related to road safety with self-driving vehicles and concludes that expectations for improved road safety may be, in some cases, overblown.

In their paper, Michael Sivak and Brandon Schoettle assessed safety from four perspectives: 1) Can self-driving vehicles compensate for contributions to crash causation by other traffic participants, as well as vehicular, roadway, and environmental factors? (2) Can all relevant inputs for computational decisions be supplied to a self-driving vehicle? (3) Can computational speed, constant vigilance, and lack of distractibility of self-driving vehicles make predictive knowledge of an experienced driver irrelevant? (4) How would road safety be influenced during the expected long transition period during which conventional and self-driving vehicles would need to interact on the road?

Contributions of other participants and factors. Sivak and Schoettle noted that not all crashes are caused by drivers. While an autonomous vehicle could compensate for some crashes caused by other traffic participants (they use the example of a drunk stepping into the roadway), and could in principle respond more quickly than a human, it might not be able to stop in time in all situations because of braking limitations.

Further, a small percentage of crashes are caused by vehicle failures. The UMTRI team sugests that there is no reason to expect that such “other” vehicle failures would be less frequent on self-driving vehicles than on conventional vehicles.

Indeed, given the complexity of the sensing hardware and of the information-processing software, it is reasonable to expect that, overall, vehicular factors would likely occur more frequently on self-driving vehicles than on conventional vehicles.

—Sivak and Schoettle

Certain roadway (e.g., floods and downed power lines) and environmental factors will also cause problems.

Information. Fully realizing autonomous driving will require detailed mapping, including locations of streetlights, stop signs, crosswalks, lane markings, and all other crucial elements. Further, this information will need to be updated in real time.

The role of experience. Fatalities per distance driven is heaving influenced by the age of the driver, and is represented by a U-shaped function with the lowest rates for middle-aged drivers.

To the extent that not all predictive knowledge gained through experience could exhaustively be programmed into a computer (or even quantified), it is not clear a priori whether computational speed, constant vigilance, and lack of distractibility of self-driving vehicles would trump the predictive experience of middle-aged drivers.

—Sivak and Schoettle

Transition period. Turning over the vehicle parc in the US takes a long time—the current average age is 11.4 years. There will be at least a several-decade-long period during which conventional and autonomous vehicles will need to interact.

One main concern during this transition period is that drivers of conventional vehicles would have certain expectations about the likely actions of other vehicles (depending on factors such as the location of the interaction, the type of the other vehicle, and the age and gender of the driver of the other vehicle, etc.). … Furthermore, in many current situations, interacting drivers of conventional vehicles make eye contact and proceed according to the feedback received from other drivers. Such feedback would be absent in interactions with self-driving vehicles. The degree of the importance of both driver expectations and feedback from other drivers, and the consequent effects on the safety of a traffic system containing both conventional and self-driving vehicles, remain to be ascertained.

—Sivak and Schoettle

From their discussion, Sivak and Schoettle drew four conclusions:

  • The expectation of zero fatalities with self-driving vehicles is not realistic.

  • It is not a foregone conclusion that a self-driving vehicle would ever perform more safely than an experienced, middle-aged driver.

  • During the transition period when conventional and self-driving vehicles would share the road, safety might actually worsen, at least for the conventional vehicles.


  • Michael Sivak, Brandon Schoettle (2015) “Road Safety With Self-Driving Vehicles: General Limitations And Road Sharing With Conventional Vehicles” UMTRI-2015-2 (UMTRI publications found here when released.)



It is not a foregone conclusion that a self-driving vehicle would ever perform more safely than an experienced, middle-aged driver.

Do I detect some ass covering here ?

But OK, if that particular demographic is to represent the benchmark human driver then I, for one, would be quite satisfied if the autonomous vehicle could perform even as safely as an experienced, middle-aged driver. And by the same token it would follow that its ability should likely surpass the skill of all lesser drivers as well. You know, those drivers responsible for the road deaths of 34,600 people last year in NA. Would you agree ?

As we like to say here - Good enough is not the enemy of perfect.


Many interesting questions here.

I think we will see a while new set of accidents from (for instance) people rear ending auto cars which obey the rules of the road slavishly. Or people who get so frustrated by driving behind a car that won't break the speed limit (even a little).

Will the programmers allow an auto car to break the speed limit, and what will happen if they do and the car crashes?
Will they let the driver set the margin by which a car breaks the speed limit? [ and so on ].

As they say, middle aged drivers are pretty good, younger and older ones are the problem. Giving an auto car to an older person is a no-brainer, but what about younger people - do you want them to learn how to drive on "manual" or not ?

I think there will be a fair few crashes by auto vehicles during the first decade or so. This will be due to unforeseen circumstances, or programming errors. These will get weeded out over time. There may be limitations on auto cars - There maybe some cities or countries where the driving is too lawless for them to work safely. (Naples ? Mumbai? )

It will be important that the insurance companies don't have to pay any more for someone killed by an "auto" than by a "manual". After a while, they will find that the auto cars are safer and they will encourage then, or refuse to give insurance to people with certain accidents (drunk, speeding, etc).

The transition period will be an interesting time ...


From my short existence, I would be very much pleased if a computer took over driving for most Americans. As a cyclist, driver, and a also a pedestrian I dont even bother to count the times another vehicle almost brought an end to my life.

If done right autonomous vehicles are easily safer than most drivers, especially when factors like poor visibility, adverse road conditions, and sudden emergency braking events happen. Crashes are sometimes unavoidable, but quick reactions, and an ever vigilant set of eyes(which are more up to the task then our own) can greatly reduce the risk of crashing and could likely lessen the chance of death.

Also, odds are the cars can react to situations better than most "experienced drivers". Computer algorithms for an event like a crash or loss of control could easily be better than most drivers, even good ones.

Snow/ice, and objects found in the roadway would be interesting situations to deal with. Stuff that falls off of trucks or blows into the pathway of the vehicle could present a very interesting challenge. I'm thinking of boxes, or even things like tumbleweeds...

But another thing to point out is, odds are these systems will be no more than a autopilot/ highly evolved cruise control... Which would still require a licensed driver behind the wheel.


Self-driving units can easily be self-checked and self calibrated (if required) at regular intervals.

Results could be stored in the on-board PC and/or transmitted to the owner.

The comments to this entry are closed.