San Diego, CA. Mary (Missy) Cummings describes herself as not a typical professor—she spent time in the “real world” as a naval officer flying F-18 fighters off aircraft carriers. She now has a Ph.D. in systems engineering and teaches at Duke University’s Pratt School of Engineering, where her interests include human-unmanned vehicle interaction and the ethical and social impact of technology.
Delivering a keynote address at IPC Apex Expo on the topic of drones, she downplayed the role of pilots, although she is one herself. An F-18 can guide itself to an aircraft-carrier landing, and in fact pilots are instructed to keep their hands off the controls when an F-18 is catapulted into flight in order to avoid pilot-induced oscillations. Pilots consider themselves the best of the best, she said, and when told to hold onto handlebars above the glare shield during takeoff, they must ask themselves, “what are we, five?”
She does believe that human pilots have a role to play on commercial airlines—controlling unruly passengers, for example. Cargo pilots may have something to worry about, however.
Bur rather than completely replacing pilots, she expects automation to help alleviate a worldwide shortage of pilots, with one pilot per cockpit assisted by robotics. She likened the approach to R2D2 occupying a spot behind Luke Skywalker in his X-wing fighter. Nevertheless, she said, the planes you fly on today are essentially drones, and their level of automation points to a bright future for drones.
A key part of her work is determining how to balance humans and automation. When assessing prospects for automation of a task, Cummings said, she makes use of the SRKE taxonomy—for skills, rules, knowledge, and expertise. The first two involve mastering skill-based reasoning and then following rules or procedures that must happen in certain sequences under different situations. Next comes knowledge-based reasoning, which, when you do it, involves running simulations in your head, employing intuition when lacking some information. Expertise comes into play in situations of maximum uncertainty—she cited aircraft captain “Sully” Sullenberger’s safe landing of an airliner in the Hudson River in January 2009 after bird strikes disabled both engines.
Computers, she said, are good at skills-based reasoning and can follow rules, better than people can. They have more trouble as uncertainty increases—self-driving cars, for example, have trouble when four arrive simultaneously at a four-way intersection. She added that if your job—such as the jobs of Apex Expo attendees—requires knowledge-based reasoning amid massive amounts of uncertainty, a robot will not replace you. Nor will robots replace maids anytime soon, she added, although robotic housekeepers are a frequent request she receives.
Railroad trains should be easy to automate, she said, as there is no lateral uncertainty. She is skeptical about self-driving cars, despite the overly optimistic enthusiasm for them in what she called California’s “Silly Valley.” They must compute everything around them and what everything around them is doing. The ultimate test case for them, she suggested, will be driving in bad weather in Boston. She also commented on autonomous vehicles operating in mines. Autonomous trucks are so precise that randomness needs to be programmed in to prevent them from forming ruts as they traverse the same path over and over. (Read more in “Autonomous mining equipment years ahead of car development” in our upcoming March issue.) Diggers, however, rely on haptic feedback still best interpreted by humans.
Drones occupy a middle territory between trains and cars, Cummings said, and show promise for applications ranging from package delivery to pipeline inspection. Remaining challenges include AI brittleness, mismatches of technology with human capability, sensor immaturity, and lack of test plans.
Cummings also cited distraction as a problem with partial automation—the Pavlovian response of a pilot or driver picking up a mobile device when becoming bored as the plane or car operates itself. In addition, partial automation can hinder human skillsets. A 2013 article in the Economist notes that younger pilots tend to trust their plane’s automation more than their own skills. Sullenberger’s background is becoming the exception rather than the rule, the article adds. He learned to fly at the age of 16, flew F-4 Phantoms, and had accumulated 40 years and 20,000 hours of mostly hands-on experience before performing “his heroics on the Hudson.”
Cummings suggested that regulation is as much an impediment to autonomous drones in the United States as technological challenges. The FAA, she said, is staffed with aviators with a romantic view of humans in the cockpit—a vision they don’t want to let go of.
Nevertheless, she concluded, the future for drones is beyond bright, and she suggested we might see Jetsons-like flying cars before we see self-driving automobiles on the streets of Boston.
