Autonomous Driving @ the Crossroads

One of the most terrifying moments one can experience as a driver or passenger in a Tesla Model S driving with autopilot turned on is the realization that the system cannot recognize intersections or traffic lights. It seems like such a basic and obvious requirement for automated driving but the Model S can’t hack it – which is one of many reasons the system is not considered an autonomous driving solution – maybe advanced cruise control is more accurate.

But the reality is even more stunning. The U.S. Department of Transportation does not have a complete inventory of U.S. intersections and traffic lights. In fact, the U.S. DOT won’t have such a complete inventory for years to come.

Talk to anyone in the traffic business and you will be told that the system for estimating how many intersections there are boils down to a rule of thumb: 1 traffic light for every 1M in population. This calculation yields somewhere between 300,000 and 330,000 intersections in the entire U.S. But it’s only an estimate. No one knows the actual total.

Now it would be logical to then consider how many of these intersections are connected to the Internet and can thereby be managed or changed dynamically to respond to evolving traffic conditions. Best guesstimate I have seen is 115,000 in the U.S. – but this, too, is only a guess.

Companies like Global Mobile Alert and ConnectedSignals among others are working on solving this riddle. Global Mobile Alert is building a system to alert drivers, distracted by their smartphones, to the proximity of traffic lights. ConnectedSignals’ Enlighten app is designed to inform the driver of the signal phase timing of upcoming traffic lights.

For now, the lack of a comprehensive intersection inventory represents a massive barrier to the advance of autonomous driving system development. If your car can’t grok the presence of an intersection, it’s going to have a tough time determining the phase of the light.

The Google car’s purely sensor-based system and its slow speed of operation is designed to recognize and cope with intersections. But the Google car is only tuned to operate in a limited and thoroughly mapped area.

The story is even worse. After all, we know that one third of all traffic fatalities occur at intersections. It would seem that intersections are a logical place to start working on whittling down the recently increasing rate of driving fatalities in the U.S. – but the relevant regulatory authority doesn’t even possess the relevant data.

So we can rest assured that autonomous driving is years away from becoming reality while we lose sleep knowing that our best minds don’t even know where all the intersections are located – let alone which ones are or are not signalized or connected. Is this any way to run a transportation system? We clearly have a lot of work ahead of us. Look out!

Roger C. Lanctot is Associate Director in the Global Automotive Practice at Strategy Analytics. More details about Strategy Analytics can be found here: https://www.strategyanalytics.com/access-services/automotive#.VuGdXfkrKUk