Engineering AVs to ‘Play Nice’ With Humans

November 1, 2018

Sam Abuelsamid

More than a decade after the DARPA Grand Challenge program demonstrated the technical viability of self-driving vehicles, the industry is now preparing its first commercial deployments of the technology at scale. While Waymo, GM Cruise, Aptiv and others begin to ramp-up automated ride-hailing services for paying passengers, unanswered questions remain. Many of them revolve around the “social” aspects of the technology.

Engineering is about taking the principles learned from science and applying them to solving real-world problems. But if the resulting technology requires people to change their behavior for it to work reliably, the engineers are doing it wrong.

The reality of automated driving is that it’s going to take many years, decades even, to largely supplant human driving. Even if AVs do take over, these vehicles are going to have to safely coexist with humans on foot, on bicycles or on scooters. If AVs can’t play nice with other road users, engineers developing them need to slow down and rethink their approach until the vehicles function properly within the overall mobility ecosystem.

This is where things get complicated. The realities of physics and of human behavior dictate that humans walking or riding a two-wheeler can change direction or intent far more rapidly than another vehicle can. And while AVs can easily communicate with each other via V2V to indicate their intent, machines communicating with humans is much more problematic.

Researchers such as John Shutko at Ford and Melissa Cefkin at Nissan have been studying how humans and AVs will interact. In 2017, Shutko’s team famously ran an experiment with the Virginia Tech Transportation Institute using a Transit Connect van equipped with a series of lights. The driver of the van wore a so-called “seat-suit” that made it look as though there was no driver in the vehicle so that the researchers could observe the responses of pedestrians to various signals.

Similarly, Cefkin and her team at Nissan have been studying the non-verbal signals given off by humans as they interact in crowded spaces, to understand if similar subtle movements by AVs could be utilized to signal their intent. Other companies, including Toyota and Drive.AI, have been testing a variety of signage to provide indications from AVs to pedestrians.

All these efforts, however, will only be successful if everyone derives the same meaning from a given signal. More than a century of motorized personal transportation has made us understand what turn signals or brake lights mean when we see them. These non-verbal messages are used globally on road vehicles and are language-agnostic. That’s important in a market like India, whose 22 major languages and many more separate dialects make any text-based approach seem unlikely.

At the recent WardsAuto UX conference in Detroit, Shutko called for the industry to come together to develop standard approaches to signaling AV operational intent. Based on the lessons learned from the 2017 experiment, the Argo AI test fleet operating in Pittsburgh, Miami and Dearborn, Mich. already has been equipped with lights to signal yield, active driving and start to go. Those involved in AV-experience design should be reaching out to Shutko to get involved in this effort.

With the explosion of micro-mobility services in 2018, including e-scooters and dockless bike sharing, it’s also time for cities to set policy around these vehicles to facilitate safe integration with AVs and pedestrians in the coming years. Consistency in this area is important.

Protected bike lanes that can be used by scooters, getting them off the sidewalks, will improve safety for all. At the same time, the addition of a physical barrier would help to keep AVs a bit more isolated from non-automotive road users. None of this comes free, of course.