Notes from the Industry: Autonomous Trucks?

The report on the fatal accident involving an autonomous test vehicle and a pedestrian (March 2018 in Arizona) came out this week. It illustrates well-documented, but under-appreciated facts of automated safety technology. The report revealed two important facts: First, the technology was unable to determine that the pedestrian involved was about to walk into the path of the car, nor was it capable of automatically stopping the vehicle if it had. Secondly, the test was dependent on the human monitor recognizing the danger and stopping the car. It is obvious that this technology was not really ‘autonomous,’ capable of driving safely itself. Limitations like that will be a general problem over the next 10 years or so as the technology wrestles with the prodigious recognition problems involved. We should say then, that such technology is a ‘driver-assist’ device rather than a ‘driver-replacement’ device. Current active cruise systems on passenger cars will follow a moving car reliably, but will not stop for a stationary car at a light. The radars involved cannot distinguish stationary cars from background clutter, but they can see moving objects. We are reminded about how much more work needs to be done before technology can mimic the extraordinary capabilities of the human mind.

Human Monitoring?
The report also revealed that the human monitor was watching a movie on her tablet at the time of the accident. This sad example reminds us that the concept of human monitoring of automated systems is a flawed process that can only work if and when problems are easily seen and diagnosed, and reaction times are measured in minutes, not seconds. If the system is truly automated, no human has the discipline, or stamina, to be actively monitoring conditions as if he or she was actively driving constantly. Put differently, what is the point of spending the money to automate if the driver still must drive the truck in his or her mind? The automated system must be fail-safe, as an elevator is. Using the example from Arizona, the car should have been stopped automatically as soon as it detected a possible threat within the danger zone. Of course, such technology today would be stopping the car, at times in panic mode, much too frequently in an urban environment. As I said before, the technology has a way to go.

Automatic Tools Can Make Us Safer
That said, there is a useful middle-ground for such ‘driver-assist’ technologies. A collection of active cruise and other collision-avoidance technologies have reduced truck rear-end collisions by 70% for some users. In this case, the ability of the sensors to reliably, continuously, monitor following distances is substantially superior to a human’s ability to do that. Fortunately, we have the matching of a workable automation (vehicle-following radar) and a known failing of human driving. Success! As long as the carriers and their drivers know what the limitations of their systems are, partial automation can work.

When I drive my well-equipped car, I know I must always be steering, watching for pedestrians and stopping for lights, stop signs, and stopped vehicles no matter what the new car ads say. So, I am now a safer driver due to the technology, but I still have to be alert, even if I can relax some under the right circumstances. Since the big benefits of automated driving depend on taking the driver out of the truck, leaving no monitor of any skill, we can see that major change is still a good ways off, at least 10 years—unless the application is in a controlled environment where a pedestrian cannot wander into a driving lane as in the Arizona tragedy.

Noël Perry is Principal with Transport Futures, located in Lebanon, PA. He can be reached at nperry@transportfutures.net.