The Last Mile

The race to autonomous driving is looking a lot less like a race these days. German automakers pushed the likely date for Level 5 autonomous driving back to 2032 from 2027, according to attendees at the International Congress for Automotive Electronics (ELIV) in Bonn last month.

There are a number of reasons for this. The first is cost. The amount of processing needed to make the split-second decisions needed to navigate through unpredictable scenarios, such as children running into the street, drivers ignoring traffic signals, or a sudden hailstorm, is well beyond the $5,000 to $7,000 price tag that carmakers have deemed affordable to drivers. There might be some wiggle room in there, but not enough to pay for several supercomputers, which is what some experts say is required to navigate through densely packed urban areas.

That cost can be reduced through massive infrastructure improvements, such as lanes restricted to autonomous vehicles. But that leads to a second problem, which is differentiation. Adding infrastructure improvements would require standardized ways of communicating with that infrastructure, which threatens the carmakers’ ability to differentiate. And if vehicles strictly adhere to speed limits and acceleration/deceleration are restricted by the infrastructure, then the differentiation between one brand and another would be largely based on the cabin interior and amenities. That comes down to electronics and seating, and it’s not clear that automotive companies are equipped to fight that battle.

A third problem involves the intermingling of human-driven and autonomous vehicles, and of pedestrians and autonomous vehicles. Today most carmakers have the technology to keep a car in its lane, even on twisty roads, which is why there were rather dubious pronouncements that autonomous driving would take over the roadways by the end of this year or next year. But as everyone who has driven a car with lane departure technology knows, vehicles can be fooled by lane shifts in construction zones.

Even the most advanced self-driving systems will not inch out into traffic when pedestrians are in a crosswalk. In a city, that can cause a massive traffic jam unless there are clearly marked turn signals everywhere, and even then pedestrians can ignore the signals and hold up traffic because they know autonomous vehicles won’t hit them.

That said, on a limited access highway with no construction, autonomous vehicles generally perform better than those driven by people. They never get distracted by texting or fall asleep. They never exhibit road rage no matter how many cars cut them off. And they can react to changes in distance and sudden braking much faster than humans. In a world with only autonomous vehicles and limited access roads, they are significantly safer and far more predictable than vehicles driven by humans.

The real challenge will be the last mile, which has always been the problem in these kinds of rollouts, and that raises the question about whether these vehicles will ever truly be able to be steering-wheel optional. Drivers always will need a way to take control of their vehicle — even if it’s only to flee from a burning forest through falling branches. There are always unexpected corner cases.

It’s time to reassess what exactly vehicles can do, what they can’t do, and how everything is going to work together as a system of systems. At this point, we’ve merely scratched the surface. While cars can navigate some roads very well, it’s a big leap to full autonomy where a car’s AI logic can react intelligently to what it wasn’t specifically trained to do.