Wednesday, December 13, 2017

Dr. Baura on Drowsy Drivers and Self-Driving Cars


Loyola’s Dr. Gail Baura spoke on sleep deprivation research in long-haul truck drivers, and its relevance to the current debate on self-driving cars.  Dr. Baura’s assigned article, “The Sleep of Long-haul Truck Drivers”, from the New England Journal of Medicine, reported on a study done in 1988 using multiple methods of data recording.  This included a questionnaire on sleep habits, electroencephalogram and eye-movement recordings, infrared videos of the driver and of the speed and position of the truck, along with polysomnography during sleep periods each night.  Together these data were designed to analyze the drowsiness of truck drivers and the effects it has on their performance.  The study used 80 drivers on normal, revenue-producing routes of two 10-hour schedules and two 13-hour schedules, each set at different times of the day. 
On average, the drivers reported a need for 6-8 hours of sleep each night to be fully alert, but over the course of this study only got 4.78 hours of measured sleep per night.  The results suggested circadian influence when incidents of measured stage-1 sleep at the wheel only occurred during 11pm and 5am.  These findings show that shift workers need to be informed of the importance of sleep and a reliable schedule that follows the body’s natural circadian rhythm, and the risks that come with sleep deprivation. 
Dr. Baura explained the problems with this study in particular and the topic as a whole.  She says there are many proposed measurements of driver drowsiness and performance, and this one is not the best, but is also not the most recent, and there is no true “winner” when it comes to this field of study.  Each method has its flaws, and it’s up to the opinion of the researcher to determine which has the best trade-offs.  This is because drowsiness is variable with each subject - some people have droopy eyes and a low heart-rate, while others will have a glassy-eyed stare or rapid blinking just before drifting into the first sleep-state.  This makes it hard to design a device that can record and recognize these symptoms, and could lead to a high rate of false alarms if they were ever implemented as safety systems in commercial and noncommercial vehicles.
We know this is true because the safety systems already implemented in noncommercial vehicles have overwhelming reports of false alarms, according to Dr. Baura, making the warnings more of an annoyance than a life-saver.  Dr. Baura said this is largely due to the problems with the sensors.  Most sensors used in manual and self-driving cars are less than ideal, because each has its own “blind-spots” - satellite, for example, is useless on a cloudy day.  Some manufacturers try to get around this by combining multiple types of sensors, as do researchers in the field of this study, but Dr. Baura emphasized that this does not mean they add up to a more accurate analysis, it just combines a lot of iffy data and gives false hope for a safer solution.  In Scientific American’s article, “Redefining ‘Safety’ for Self-Driving Cars”, this is an example of how self-driving cars are far from perfect. 
This problem is rooted in the main obstacle for self-driving cars today - human error.  We have yet to find a perfect solution to analyzing a car’s surroundings without risking false assumptions, and poor sensors are only the beginning of the problem.  Even when the right assumption is made, the car has to be programmed to find the safest solution to the environment.  Human-error has been the sole source of self-driving car accidents, aside from one incident in which a truck backed up and hit a self-driving shuttle that had sensed the truck but was only programmed to stop and wait.  The truck hit the shuttle’s bumper and then stopped, so no serious injuries occurred.  However, this exposes the greater issue of unpredictable human error and the infinite ways it can out passengers in danger.  Self-driving cars must obey traffic laws, but car accidents usually take place when drivers disobey these laws, and the safest response is often to disobey laws as well.  The article explains that when a situation does not have a predetermined response, self-driving cars will pull over and stop until the environment returns to normal.  This sounds logical, but Scientific American argues that this is not always the right choice.  What if the best response is to speed up and avoid a collision, or to swerve before there’s time to signal? 
In both drowsiness research and self-driving programming, the biggest obstacle is creating a system that accounts for all human variability, but human behavior is largely unpredictable.  Until we can find a way to program for this, self-driving cars will never be 100% accident-free while sharing the road with human drivers - drowsy or not.

Sources
Mitler, M. M., et. al. (1997) The Sleep of Long-Haul Truck Drivers. Massachusetts     Medical Society, The New England Journal of Medicine, 24 Feb. 2016, https://luc.app.box.com/v/neuroseminar/file/251218239087
Saripalli, Srikanth. “Redefining ‘Safety’ for Self-Driving Cars.” Scientific American, The Conversation US, Inc., 29 Nov. 2017, www.scientificamerican.com/article/redefining-ldquo-safety-rdquo-for-self-driving-cars/.

No comments:

Post a Comment