October 30, 2015
Driving around with Steve Smith, it's sometimes unclear whether he's a guy that fixes traffic problems or causes them.
“This is the classic Pittsburgh left turn,” Smith says, laughing, waiting for his chance to bolt—illegally—in front of oncoming traffic the moment the light turns green. “Generally, it’s an accepted practice in Pittsburgh if you’re at a light, and want to turn left, that approaching vehicles expect you to dart in front of them. "
Smith is only half joking when he insists this practice of "squeezing one guy through" helps traffic flow better. But traffic flow is something he knows a lot about. A professor in Carnegie Mellon University's robotics department, he’s the guy behind the roll-out of an experimental adaptive traffic signal system meant to heal traffic congestion in the city’s busy East End.
“It operates totally decentralized, so each intersection watches the traffic that’s coming in all directions," Smith says. "And then in real time, it builds a plan so that all the vehicles it sees move through the intersection in the most efficient way possible.”
That means a computer algorithm essentially figures out how long to leave the green lights green and the red lights red in order to maximize traffic flow in all directions. Steve’s lights are also all talking to each other, so neighboring signals can coordinate their pattern of green and red to keep the flow going for blocks at at time. As proof, Steve drives through the East Liberty test zone near the intersection of Penn Avenue and Centre Ave and makes it through five lights in a row. Once outside the reach of his smart lights— right on cue—his river of green abruptly turns red.
“I guess it’s back to reality,” he says, laughing.
Figuring out how long to leave lights green or red—and then coordinating them all into one well-tuned system—is usually something traffic planners do. And they have to do big expensive traffic studies and construct elaborate plans to deal with all the city’s problem areas on a case by case basis. The beauty of Smith’s system is that there is no plan. Or more accurately, there’s a computer doing a traffic study and making a plan—and tweaking it—continuously.
“We’re adjusting second by second, cycle by cycle. Our system kind of discovers what the dominant flows are how they change throughout the day. And the system is self-tuning. It controls itself.”
In fact, Smith and his colleagues hardly ever make manual adjustments to the system, which can easily react and adapt to unexpected events like accidents or a parked delivery van temporarily occupying an entire lane. When things like this happen, the system just sees it, figures it out and adapts. And this is a lot more effective than a traditional traffic plan.
“Consistently, we get 25 percent reduction in travel times; over 40 percent reduction in waiting times. With respect to air quality, we’ve reduced emissions on the order of 21 percent.”
Up next, Smith and his team are tackling Pittsburgh’s North Shore—home to two big sports stadiums and regular game day traffic gridlock. After that, he wants to start pushing the systems out to other cities.
“We’re still working on the pricing model, but we’re thinking they’ll cost about $20,000 per intersection.”
That might sound like a lot. But in the world of traffic technology, it’s actually pretty cheap. Cameras alone can cost $5,000 a piece. And Smith’s technology plays well with a variety of signal systems, which should make the adaptive technology basically a plug-and-play add-on for most cities.
He’s also working on ways to make pedestrian crosswalk signals interact with traffic lights. And he wants to fix one of his biggest pet peeves: Those big yellow call buttons at crosswalks that don't do what you think they do.
“If you push the button, it doesn’t get you there any faster. But what happens is, if you do push it, it will make sure it gives you enough green to make it across the street safely.”
In Steve’s new version of the system, those call buttons could actually make your waits shorter. And probably not too long from now, those crosswalk signals will just already know you’re there.