This AI traffic system in Pittsburgh cut travel time by 25%
Pittsburgh drivers add 81 hours of overtime to their trips each year due to traffic, according to a TomTom survey. Although there are other American cities who have worse, Pittsburgh is known for its difficult driving conditions, with hills, bridges and cyclists, all in a city without a network where many intersections are not allowed on red signs. But Pittsburgh drivers may soon be relieved.
The varied road conditions make it difficult to get around, but that’s also why companies like Uber come to Pittsburgh to test autonomous vehicles. If traffic technology can work in Pittsburgh, it can work almost anywhere. And, with AV, this traffic technology includes Surtrac, an AI system that allows traffic lights to adapt to traffic conditions instead of relying on pre-programmed cycles.
At the lights where Surtrac is installed, the team behind the system estimates that it has reduced travel time by 25%, braking by 30% and idling by over 40%. It costs about $ 20,000 to wire and install Surtrac at an intersection.
Surtrac works by detecting traffic and creating predictive models. First, equipment, including a computer, camera or radar, is installed at the intersection. Surtrac can then see the cars arriving at the intersection from all directions. The computer runs a predictive model and uses it to generate a real-time signal timing plan. Processing is done in a way that, through communication with downstream models, builds a local plan from multiple data sources.
Each intersection controls its own traffic, but by communicating the projected outgoing flows to neighboring intersections, those intersections can better prepare for incoming traffic.
Surtrac, which started as a project at Carnegie Mellon, flew to 12 high-volume intersections in 2012. It is now 50 intersections with 150 more on the way, paid for with a grant from the Federal Highway Administration. In 2015, the project was born out of Carnegie Mellon as a company called Fast Throughput Technologies.
After the pilot, Steve Smith, professor of robotics at Carnegie Mellon and director of Rapid Flow Technologies, said they could notice a significant difference in traffic. But they were quickly informed that they had forgotten about non-motorized traffic.
“We immediately received a lot of feedback from pedestrians, who felt left out of the image,” Smith said.
Changes to the system have resulted in a maximum waiting time for pedestrians at traffic lights. Carnegie Mellon researchers and students are working on a side project to create a mobile phone app to communicate with lights for people with disabilities who need longer to cross the street.
“It has been said that solving urban traffic can be more difficult than sending a rocket to the moon.”
Director, Adaptive Traffic Management and Operations Laboratory
The system is fully automated, but can be installed in real time at a central location if desired or required. Smith said, however, that they didn’t really expect people to intervene manually.
“In theory, this is one of the best,” said Alexandre stevanovic, Associate Professor in the Department of Civil, Environmental and Geomatics Engineering at Florida Atlantic University and Director of the Lab for Adaptive Traffic Operations & Management (LATOM).
Stevanovic said it was still a “theory” as it requires more testing, namely at least half a dozen more sites with different traffic patterns, like longer blocks with more traffic. fast. But he commends Surtrac for looking at previous technologies and collecting as much information as possible.
“There is nothing wrong with needing improvements, these are complex systems,” said Stevanovic. “It has been said that solving urban traffic can be more difficult than sending a rocket to the moon.”
Surtrac is expanding beyond Pittsburgh – even beyond Pennsylvania – this year. He goes to 25 intersections in Atlanta and 15 in Beverly Hills. King County, outside of Chicago, is also in line for the Surtrac deployment.
Ultimately, Surtrac will operate with autonomous vehicles. Smith said they had worked in recent years for traffic light control with connected cars, noting that he wanted the system to be prepared “for that eventuality.” A recent study found that having commercial vehicles on the road was another factor in improving traffic.
Traffic control could be even better when information flows between infrastructure and cars. In a simulation, Smith was able to show whether a vehicle is ready to share its route with the intersection, such as with dedicated Short Range Communication Radios (DSRC) or a navigation device, vehicles move on the network 20% faster without affecting non-equipped vehicles.
“It sounds like magic,” Smith said. “But once the world is connected, we’ll always know where the cars are.”
Smith said they were investigating whether Surtrac could one day detect traffic accidents and other events in real time, so they could start using the information to offer rerouting advice to vehicles. They are also exploring different machine learning algorithms to reduce some uncertainties related to sensor data.
Although Surtrac will be at 200 intersections in the near future, there are over 600 intersections in Pittsburgh. Smith said they haven’t noticed a plateau in improvements as they expand – so traffic may one day be a thing of the past in Pittsburgh.
“I feel like the more network you can encompass, the smoother you will travel,” he said.