Michael Sivak and Brandon Schoettle of the University of Michigan Transportation Research Institute (UMTRI) are suggesting that self-driving vehicles should be required to pass a licensing test.
Most driver-licensing tests evaluate three aspects of driving readiness, they observed: visual performance; knowledge of rules and regulations related to driving and traffic in general; and driving-related psychomotor skills. The reasons for the need to test prospective drivers are that:
Vision is essential for driving;
Driving and traffic laws and regulations need to be learned and followed;
Driving-related skills need to be acquired through practice; and
People differ in performance for all of the above aspects.
In a new UMTRI white paper, they suggest a number arguments in support of their proposal for testing autonomous vehicle as well:
Sensing hardware, spatial maps, and software algorithms will vary among manufacturers of self-driving vehicles, resulting in variability of on-road performance—as is the case with humans.
Visual and sensing performance of self-driving vehicles in inclement weather is not yet sufficient.
Visual-pattern recognition is a potential problem for current sensing systems in self-driving vehicles.
An example of a difficult pattern-recognition task for a computer: Flooded roadway. (Photo: © Chris Talbot.) Source: Sivak and Schoettle. Click to enlarge.
Current self-driving vehicles have not yet been tested thoroughly under a variety of demanding conditions (e.g., in snow).
On-road performance of some current self-driving vehicles is not yet perfect, even in good weather.
Self-driving vehicles will face, on rare occasions, ethical dilemmas in their decision-making.
For self-driving vehicles, Sivak and Schoettle observe, experience under one set of conditions that requires certain hardware or software capabilities does not improve performance under a different set of conditions that requires different hardware or software capabilities. This is in contrast to novice human drivers.
Thus, they suggest, the underlying logic for the use of graduated driver licensing (GDL) systems used for novice young drivers does not apply to self-driving vehicles: A self- driving vehicle either has the hardware and software to deal with a particular situation, or it does not. If it does not, experience in other situations will not be of benefit.
On the other hand, the GDL approach would be applicable should a manufacturer explicitly decide to limit the operation of its vehicles to certain conditions, until improved hardware or software become available. For example, a manufacturer might feel confident that its vehicles could handle all situations except nighttime and snow. In such a situation, after passing a licensing test related to the limited conditions, the vehicle would be given a provisional license that would exclude nighttime driving and driving in snow. A full license could then be obtained once future updates to hardware or software are developed and made available, and the updated vehicle passes an unrestricted licensing test.—Sivak and Schoettle (2015)
Michael Sivak and Brandon Schoettle (2015) “Should We Require Licensing Tests and Graduated Licensing for Self-Driving Vehicles?” UMTRI-2015-33