Ann Arbor Area Business Monthly
Small Business and the Internet
Autos On Full Auto
By Mike Gould
Baby you can drive my car, and maybe I'll love you.
Beep beep'm beep beep yeah
The Beatles: Rubber Soul, Lennon–McCartney, 1965
In the spirit of last month’s alarming speculations about Artificial Intelligence (AI – URL below), self-driving cars are currently in heavy rotation, media-wise. Driverless cars don’t have much in the way of AI (yet), but they have enough to be dangerous, if last week’s first death-by-Uber is any indication.
To re-cap, a woman in Arizona was killed by an autonomous Uber car while crossing a busy street in Tempe. First reports blamed the victim, who was not in a crosswalk, and was pushing a bicycle after dark. Analysis is continuing as this is being written, but it is starting to look like a failure of the car’s sensors or computer, or a combination of both was at fault. There was a human riding shotgun in the car, but she was not watching the road, which indicates a failure of Uber’s training of personnel.
I Say LIDAR, You Say LASER
Velodyne, the manufacturer of the LIDAR sensing system involved, is saying that their gear works perfectly fine in the dark and are at a loss as to the cause of the accident. LIDAR which stands for, um, Light RADAR, sorta, is a means of using invisible infra-red laser beams to detect obstacles, the sides of the road, and surrounding traffic. This device is mounted on the roof of a car, scanning about as the car moves. It reports what it “sees” to an on-board computer, which builds an internal picture of the car’s environs. It then uses this map to aim the car and hopefully avoid crashing into things.
The smarts built into the computer are collectively referred to as an “Expert System”, a set of rules describing how to recognize obstacles, and how to avoid them. As the cars are driven around test tracks, and now, city streets, the expert systems increases their knowledge of traffic patterns and behaviors, theoretically becoming better and better drivers. Same with humans; the longer you drive, the better equipped you are to respond to emergencies.
In Ann Arbor, we have Mcity, a test environment on the north side of town where a model community is set up to test self-driving cars in a safe area devoid of potential casualties. Uber decided that it is better to test things in real-life situations on the streets of cities in Arizona. Why there? Because Arizona has no regulations regarding turning their streets into test beds for robot cars. Uber got kicked out of testing in California, and headed for the more corporate-friendly roads of Phoenix.
I Welcome Our Robot
Mostly. A lot of issues here, and safety is only one of them. Let’s talk the positives first. I hate to drive, and will do my best to get whomever I am traveling with to get behind the wheel.
There are also a lot of people with disabilities who can’t drive and would welcome a better means of getting out of their houses. And once all vehicles are LIDAR-equipped and talking to each other and to Traffic Central via WiFi or whatever, all indications are that there will be a massive decline in traffic deaths, as humans are notoriously iffy drivers, with skill ranges going from dangerous dopers to Mario Andretti-wannabees moving at high speeds. The kicker here is that everybody has to be in a linked vehicle; a single drunk can continue to ruin your day if his/her car isn’t on the same page as every other commuter’s.
Issues With Giving Robbie The Keys
(The above references Robbie the Robot from the movie Forbidden Planet, who had his own car. A classic, see it when you can.)
First, car robots will need vastly improved brains. This whole business is still in its infancy and there are a lot of things that need to be more developed. Cars will need some form of AI to successfully recognize the difference between a toddler and a bit of road kill which can be successfully run over if necessary on a crowded road. And if a decision has to be made whether or not to swerve into a ditch in case of an emergency situation, how is that to be handled? Humans have problems with scenarios like that.
Second, computer security of these systems will need to be hardened, especially if all the cars are communicating with each other. What if a state-sponsored villain mounted an attack on our traffic system, hacked into our cars’ computers and ordered them to smash into each other? One virus-infected self-driver could infect an entire rush hour with catastrophic results. Recent work at the UM includes the Mcity Threat Identification Model, software that can “…help academic and industry researchers analyze the likelihood and severity of potential threats [to self-driving cars]” (URL below).
Third, cost is still a very big deal. The LIDAR systems currently in use on the roofs of test cars cost thousands of dollars. True, they get cheaper every year as the technology improves, but things will have to commodify quickly if we are to get to driving robots any time soon. Michigan is a hotbed of this research; Osram’s LIDAR division, headquartered in Novi, is very deep into working with GM, et al, to get this figured out. And of course the University of Michigan’s Mcity is a hub of this kind of research.
Fourth, is it really in the interest of our cities to continue the traffic, smog, and parking problems brought about by millions of cars, self-driving or not? Might a better solution be self-driving buses and trollies, bringing people into the city from parking lots on the edges of towns, where their self-driving cars brought them?
Whatever the road we choose, it’s going to be a bumpy one. Stay tuned, and buckle up!
Mcity - https://mcity.umich.edu
Mike Gould always looks both ways when crossing the street. He was a mouse wrangler for the U of M for 20 years, runs the MondoDyne Macintosh Training/Photography mega-mall, is a laser artist, directs the Illuminatus Lasers, and welcomes comments addressed to firstname.lastname@example.org.