Background: Beyond obviously unsafe behaviors, many minor hazards affect roadway safety. For example, a car rapidly changing lanes creates dangers for the cars it cuts off, which can be mitigated by other cars that make extra room in their lanes. And conservative driving is not always good driving — sometimes it's the dawdling or timid vehicle that causes problems.
The challenge for self-driving vehicles is that being a good driver requires sophisticated decision-making and a basic ability to understand other drivers' future actions. At the simplest level, it means not rapidly stopping when a traffic light turns red if there’s a fast-moving car close behind. In more complex cases, it also means understanding that a pedestrian entering a crosswalk is waiting for the car to cross, even though the pedestrian has the right of way.
What to watch: The development of “good driving” technology is the most important, and most exciting, area of development for self-driving cars. An autonomous car must be able to see far enough (and at high enough resolution) to understand what other road users are doing, and so emerging sensor technologies will be important — especially at higher speeds.
However, decision-making technology is arguably even more critical. Many self-driving vehicles seem to drive poorly (getting rear-ended, frustrating other road users) even at very low speeds, where current sensor performance is already adequate. This suggests that decision-making, not sensing, is the greatest bottleneck for self-driving vehicles.
The bottom line: Self-driving vehicle companies should not be adding more “so-so drivers” to the road, which makes driving more dangerous for everyone. AV companies should instead work to develop good drivers that not only avoid major accidents, but also make the roads safer by understanding how their behavior will interact with others.
Edwin Olson is the CEO of May Mobility and an associate professor of computer science at the University of Michigan.
