IBM takes traffic-jam-prediction technology for test drive

  • 13-Jun-2011 08:38 EDT
SmarterTraveler Multiple Routes.jpg

IBM Research is developing a prediction tool that alerts and diverts drivers when it pinpoints new traffic jams on expected travel routes.

No matter how rapidly highway construction proceeds, traffic congestion is bound to worsen as ever greater numbers of vehicles hit the road, so transport engineers worldwide are working on ways to speed motorists to their destinations despite increased overcrowding. Experts believe that a key remedy to the problem could be intelligent traffic-monitoring networks that would identify trouble spots in real time and route drivers around them.

Together with the California Department of Transportation (Caltrans) and the California Center for Innovative Transportation (CCIT) at the University of California, Berkeley, engineers at IBM Research are working to develop a system that will help commuters avoid rush-hour congestion while enabling transportation agencies to better predict and manage traffic flow. Company researchers have created a digital modeling tool that can collect and analyze vehicle rate and flow data, and predict where jams will occur as much as a half-hour before they would be reported by radio or other means, potentially enabling motorists to access personalized travel recommendations even before they get in their vehicles.

“Traffic studies suggest that for every minute a road incident remains in place, it creates about four minutes of delay for the following traffic,” said Greg Larson, Chief of the Office of Traffic Operations Research at Caltrans. “With this kind of system, we could detect incidents faster, which could help us to respond more quickly and so address these problems before they pile up,” he said. “Right now, we rely on 911 calls,” a process that inevitably entails delays.

Spanning the San Francisco Bay Area, the Smarter Traveler Research Initiative acquires and analyzes traffic data generated from sensors in roads, toll booths, bridges, and intersections. The system combines that information with real-time location data sent to it by GPS sensors in drivers’ mobile phones, which enables it to learn their preferred travel days and routes. Alerts are then automatically delivered via e-mail or text messages that describe the status of the motorist’s typical commuting route beforehand, thus eliminating potential driver distraction once they take the wheel.

Three-part system

The system comprises three parts, according to John Day, Program Manager for IBM Smarter Traveler, the business unit that created the prediction tool. The first is a user base of motorists with GPS-enabled smart phones who opt in to report their locations on the road network automatically, which allows “the system to learn the routes that are important to the individual. The system does this by looking at the travel end points that seem important to people,” Day explained, adding that “we don’t want to ask the users for anything that we can derive.”

The system would also gather traffic data from road sensors, such as the thousands of inductive loop detectors—magnetic induction devices that sense and report the passage of vehicles every 30 s—that are embedded throughout Bay Area roadways.

The second component is IBM’s Traffic Prediction Tool (TPT), a learning and analytics engine that employs historical data to predict new events in real time.

“It takes the traffic data and identifies correlations between small road incidents and much larger ones,” Day explained. “The TPT finds anomalies as they’re happening” and rapidly determines what traffic patterns are likely to result. Such tools, which are also a focus of researchers at Berkeley’s CCIT, typically rely on what mathematicians call Lagrangian sensing, a method by which algorithms provide optimized solutions to complex arrays of partial differential equations that represent traffic flows.

The third part of the system, Day said, involves getting the travel recommendations “out to the subscribers when the TPT sees a match between where an individual seems to want to go and problems on the projected route.”

At the same time, the system would help owners and operators of transportation systems better predict and reduce bumper-to-bumper traffic before congestion occurs through improved traffic-signal timing, ramp metering, and route planning, Caltrans' Larson said.

Mobile millennium

The impetus for the current initiative, Larson said, came from Mobile Century, a one-day, “ground-truthing” test in 2008 in which drivers of a hundred Renault cars carrying GPS-equipped cell phones reported their locations as they motored down Interstate 880 during a 12-h period. Their locations were collected and matched up against instrumented roadway sensor data to validate the concept.

“Based on those results, Caltrans and CCIT moved to Mobile Millennium [in late 2008], a more extensive project where we worked with the phone supplier, Nokia, and the traffic-information provider, Navteq,” Larson said. Participants agreed to provide GPS locations via cell phone, and the streaming data from both mobile and static sensors was consolidated in real time. The users then received in return a speed map (with routes marked green, yellow, or red) via a smart phone app.

“With as few as 10 percent of the vehicles on a road participating, you can get a very good idea of what’s going on,” said Larson.

Today, the resulting network integrates dozens of data feeds totaling millions of data points every day to produce real-time estimates of traffic on highways and major arterial roads in northern California. He noted that others, including Google Maps and Ford, are pursuing similar operations now.

In late 2009, “IBM came to us because they needed a steady source of data for their prediction tool,” Larson recalled. “We partnered them with our Berkeley colleagues.”

Although “the latest system might allow us to move away from sensing,” Caltrans researchers are now also looking at road sensors other than inductive loop detectors “since we don’t like to tear up the pavement,” he said. These systems include side-fire radars, which are easily installed but can be occluded by trucks or confused by background returns, as well as battery-powered, wireless magnetometers the size of hockey pucks that installers can just drop into holes drilled into the pavement.

The new traffic network may also eventually enable variable speed limits, where speed limits change depending on road, traffic, and weather conditions to improve flow and road safety. Variable speed limits, Larson noted, “which today are used primarily in Europe, depend on drivers agreeing to obey the posted speed. But since it is “hard for drivers to stay on the posted speed, we’re working to link to a vehicle’s adaptive cruise control, which could be set according to the speed beamed to it by a traffic monitoring system.”

The potential addition of cooperative vehicle-to-vehicle (V2V) communications, he concluded, might even “allow us to safely close the gaps between vehicles to boost throughput substantially.”

Share
HTML for Linking to Page
Page URL
Grade
Rate It
4.00 Avg. Rating

Read More Articles On

2017-03-04
Euro NCAP will establish a separate category for autonomous vehicles, but there is not likely to be one for cars that are claimed to protect all occupants from serious injury or death.
2017-01-07
Motion sickness in autonomous vehicles is the new "elephant in the room," with engineers suffering during autonomous-driving simulator runs. Researchers are working to solve this nasty issue.
2017-01-13
Range anxiety is not just affecting EV drivers on the road; it is also a significant hurdle for Formula E teams on the track. U.K. simulator specialist rFpro says its technology can help.
2017-01-09
CEO John Krafcik told the Automobili-D audience in Detroit that Waymo is building its own hardware suite with a fully top-to-bottom, full-stack approach. The classic auto industry vertical integration includes all vision sensors, radars and LiDAR, along with related “AI compute” artificial-intelligence platform.

Related Items

Technical Paper / Journal Article
2010-10-25
Training / Education
2013-04-09
Training / Education
2010-03-15
Training / Education
2010-03-15
Technical Paper / Journal Article
2010-09-28