Moving about a city not only drains valuable time and energy, but also contributes a large share of greenhouse gas emissions. Estimates show that the United States economy loses over $120 billion a year due to traffic congestion and that this congestion contributes to 27 percent of the country’s carbon dioxide emissions. The Center for Economics and Business Research predicts that these costs will rise 50 percent by 2030.
Urban mobility systems are largely inefficient today because cities are unable to dynamically route vehicles and coordinate other resources on the road in response to changing traffic patterns.Three present and persistent trends across the world today — an increasing rate of environmental damage, the accelerated movement of populations to urban areas, and exponential growth in the data that we create and consume — present a unique challenge for cities: how can they efficiently and sustainably allocate traffic using data?
The collection and analysis of traffic data can surface vehicle movement and road usage patterns over time for cities. As these patterns become apparent, cities can make strategic decisions on how to coordinate resources (e.g. traffic lights, public transit stops, parking) in order to reduce overall congestion and increase vehicle efficiency and safety in urban environments. Moreover, after enough data is collected and analyzed, cities can integrate machine learning techniques into their systems to make these strategic decisions in real-time. Today, cities are leveraging existing data and generating new data in order to improve urban mobility.
Leveraging existing data for navigation
With a wealth of data offered by traffic monitoring services, cities can now look outside of traditional departmental datasets to determine where, when, and how people and vehicles move through the urban landscape. Thanks to Application Programming Interfaces (APIs) – contracts between the providers and consumers of a service that create a common language for two nodes to communicate – real-time traffic data from services such as Google Maps, Waze and Strava can be combined with public datasets such as bus stop locations or parking to inform the routing of resources across a network.
The “Quantum Team” from MIT's Operations Research Center (ORC) in partnership with Boston Public Schools (BPS) is applying this approach today to route public school buses. MIT Professor Professor Dimitris Bertsimas and PhD students Arthur Delarue and Sebastien Martin developed a routing solution that uses both traffic pattern data from Google Maps and student address data from Boston Public Schools to determine the optimal number of bus routes, location of bus stops, and students per bus for the entire district. This approach involved clustering groups of students across bus stops, using Google Maps for traffic volume data, combining stops into various efficient routes for each school, then combining routes across schools to find the optimal route solution for the overall system. Compared to BPS’s previous methods of manually building school routes—a process that would take several weeks—the Quantum Team’s solution completes the task in approximately thirty minutes.
Analyzing this data using mapping software and optimization techniques to plan routes can dramatically reduce transportation costs. BPS reported that this new method could save the school system as much as $5 million annually by eliminating at least 50 bus routes, reduce carbon emissions by 20,000 pounds a day, and remove 1 million miles from bus trips. The new route schedule was set to go live for this school year. While results from this new routing approach haven’t been reported yet, BPS’s transportation staff have vetted the routes. John Hanlon, Chief of Operations at BPS stated that “the average walk-to-stop per student is the same as last year, as is the average commute time.” The analytics team in the Office of Performance and Accountability for the City of New Orleans now references the BPS work for employing a similar method to help improve Emergency Medical Services (EMS) response times.
Generating new data from infrastructure
Cities aren’t limited to using existing private-sector datasets to make resource allocation decisions. In fact, several public-private collaborations are emerging to create new data sources that provide departments with specific information on infrastructure use. These new datasets can not only be used to inform routing strategy today, but through machine learning techniques, can make dynamic recommendations over time.
For example, Moreno Valley, one of California’s fastest-growing cities, partnered with Hitachi Visualization to set up a video system of more than 430 cameras to design solutions for a number of the city’s challenges. The impetus for the project was to introduce efficiencies to the police department and reduce the department’s need to hire additional staff. With police department budgets and buy-in, the city decided to invest in a camera system that could serve the police department with the potential to help other departments as well. After 12 meetings over several months with different members of the Moreno Valley community, the city implemented the system.
Today, one of the primary use cases for the cameras is to optimize traffic flow at city intersections. The city set up three cameras at intersections and parks throughout Moreno Valley. Using information from this video feed, the city’s transportation management center changes lights during peak traffic times so that cars on busier routes do not stay idle for as long, thereby easing traffic flow.
In addition to the police and transportation teams, the parks, maintenance, and emergency management departments have all leveraged the system to help determine when certain resources need to be replaced or if someone reports an incident. The cameras have led to better use of city staff time, allowing employees to visualize the condition of infrastructure remotely rather than going to particular locations, investigating situations in person, and manually collecting data.
In addition to deploying cameras, cities have attached Internet of Things (IoT) to key pieces of infrastructure such as traffic lights to reduce idle time and congestion. Smart traffic lights can minimize the delay that vehicles experience by taking all of the surrounding traffic into account and determining the optimal waiting times and allocation of the road.
According to DOT, only 3 percent of the country’s traffic signals today are adaptive (i.e., make adjustments in real time), illustrating an emerging opportunity to make better use of current infrastructure to generate data for mobility resource optimization. The City of Pittsburgh has partnered with Rapid Flow Technologies and their product Surtrac to employ artificial intelligence techniques across a network of smart traffic lights so that they make real-time signal changes in response to traffic. On average, the method decreases wait time by 40 percent, decreases travel time by 25 percent, and increases average speeds by 34 percent, according to a Pittsburgh pilot study.
Specifically, Surtrac uses a dynamic programming search algorithm to find the optimal number and sequence of vehicles on the road and determine how long each green light should last based on that order. The Surtrac team also employs probabilistic modeling to estimate vehicles’ current travel state (e.g., location, velocity) and patterns (e.g., end destination), as these variables affect overall traffic flow. The Surtrac system is distributed, meaning that each intersection has its own computer that stores the data, does calculations and then communicates that data to the computers at nearby intersections, which determine the length of lights.
Flow software engineer Allen Hawkes said, “the biggest benefit of Surtrac is that it optimizes for all vehicles approaching it, multiple times a second. In other words, It’s not just learning some averages as they shift throughout the day, but rather each new green light is based entirely on each vehicle approaching the intersection.” Today, this allows for the technology to react to instantaneous changes in traffic patterns that are common in urban settings. Surtrac is operational at 74 lights across Pittsburgh and Atlanta with plans to expand to Beverly Hills, CA; Portland, ME; and Kane County, IL.
Moving towards an autonomous future
Leveraging existing data and building a smarter mobility infrastructure to generate new data help cities lay the foundation for a connected and autonomous future. As self-driving vehicle technology continues to develop, these vehicles could use sensors to directly connect and communicate with one another and with city infrastructure to optimize the traffic flow at any intersection, dynamically pay tolls, or find parking. A smart network of connected machines generating data ultimately leads to a more productive, more sustainable, and healthier city.