Congestion charges or vehicle limits as well as the phasing out of diesel engines have contributed to air pollution decreases. Now, intelligent systems are using data to provide networked traffic systems that manage traffic flows to ease the daily commute for road and pavement users.
One of the longest running smart transport projects is in Pittsburgh, Pennsylvania, USA. In July 2012, the Robotics Institute at Carnegie Mellon University deployed its adaptive traffic signal technology, Surtrac in nine traffic junction sites in the East Liberty area of the city.
The AI/robotic system treats traffic control at these junctions as a single machine scheduling issue. It senses its surroundings and what is going on in real-time at the junction based on data from a software integration/API (application programming interface), cameras, radar and even induction loops from the surrounding infrastructure. It processes the information and creates an optimisation plan every second to management traffic queues through the intersection with minimal stops for vehicles and based on the traffic volume.
A second API sends these commands to the controller to co-ordinate the traffic signals to affect this plan. Surtrac also communicates information about the strategy and the traffic flow to nearby intersections, so that it can be incorporated into its own optimisation plan. The result, said Rapid Flow Technologies, the spin-out company created to commercialise Surtrac technology, is that traffic junctions have both autonomy and co-ordinated control across the network, responding to traffic conditions.
AI manages traffic
Stephen F Smith, research professor and director of the Intelligent Coordination and Logistics Lab, The Robotics Institute, Carnegie Mellon University explained: “At the beginning of each planning cycle, a given intersection perceives the approaching (or already queued) traffic from its local sensors and builds a prediction of when it expects each approaching vehicle to arrive at the intersection. Then, in real-time it constructs a "signal timing plan” (a schedule of ‘green’ times for each intersection phase) that moves all of the sensed traffic through the intersection in a way that minimises cumulative wait time.
“Once the signal timing plan (or phase schedule) is generated, it is sent to the hardware controller at the intersection (the actual device that physically controls the traffic signal) to begin executing it. Simultaneously, the intersection sends to its downstream neighbours an estimation of what traffic it expects to be sending their way (according to the plan). Each downstream intersection is doing the same thing - generating its local signal timing plan, but now it has an expectation of what traffic is coming down the pike after the traffic it sees through its local sensors, and hence it can build a longer horizon plan.
“The planning cycle is repeated by each intersection every couple of seconds, allowing adjacent intersections to exploit shared information to converge to co-ordinated network level plans.”
In this way, the management plan can adapt quickly to congestion or capacity changes because it is based on real-time, not historical data. It also avoids creating congestion in other parts of the city with real-time updates about queues and traffic build-up from downstream intersections.
Communication is via a private message protocol with messages transmitted through TCP/IP network protocols. “Many urban centres have installed fibre optic cable between intersections, which is clearly the most efficient alternative,” said Smith. “If fibre optic cable is not present, point-to-point (line of sight) communication radios can also be used. Use of a cellular (or internet connection) is typically used as a backhaul of execution streams for purposes of computing performance statistics, etc. But all planning and execution is done through computing and communication at the edge,” he explained. The local processors run LINUX which also provides the underlying security mechanism.
Since 2012, the number of installations has increased. The initial nine junction installations doubled to 18 the following year, and 32 were added in the next two years. The city of Pittsburgh has received federal and state funding to add another 150 intersections over the next three year, covering about one third of the city’s signal junctions. These systems will manage peak traffic flows of around 4,000 vehicles/hour and an average 45,000 vehicles per day.
AI’s vision for traffic management
AI has also been used in a traffic control pilot scheme as part of the AI4Cities project, in Helsinki, Finland this summer. The scheme was developed by Marshall AI and Dynniq Finland.
Computer vision detects vehicles, cyclist and pedestrians at intersections and when the lights are green but there are no road users approaching, the traffic lights are changed to green to reduce the amount of time spent idling at the lights.
Road traffic accounts for more than a quarter of emissions in the EU and the majority of urban traffic emissions are generated when vehicles stop and accelerate at road intersections.
“The city of Helsinki wants to be carbon-neutral by 2030, and this project will contribute to this objective,” commented Kaisa Sibelius, project manager for Forum Virium Helsinki which is co-ordinating the AI4Cities project in the city.
There is a parallel pilot in Paris where visual analysis takes place at the network edge, in a control cabinet; Helsinki uses a private cloud infrastructure.
In Helsinki, the visual data is generated by nine cameras, monitoring nine incoming vehicle directions and seven zebra crossings for light traffics across three intersections. In Paris, there are 10 cameras covering three intersections (10 incoming vehicle lanes and 10 zebra crossings).
“The visual data is turned into traffic user detections, which are tracked over time and mapped to sensing areas for each intersection,” explained Nordström. “This way the system generates real-time information about demand at all times for the traffic controller, which makes sure that the traffic lights behave accordingly.
“In both trials we use existing traffic controllers, so only the cameras, data communication, processing resources and software have been added. All of these are provided by MarshallAI in both cities. The processing is based on Nvidia’s hardware accelerated graphic processing units (GPU).”
Early data from the five intersections that has been studied in detail shows that current emission levels can be reduced by between 4% and 8%, reported AI4Cities.
The current trials run at least until the end of September in Helsinki and the end of October in the Paris suburb of Meudon. After that, the cities may choose to extend the trials or integrate them permanently into those traffic intersections.
Part of the AI4Cities project’s business plan is to scale up and find new customers, said Nordström. “The main market will be Europe, but there is no reason why we could not provide the solution globally . . . likely [to] be based on local partners in each market,” he confirmed.
Mapping pollution
Emissions from traffic were also a concern for Romain Lacombe, founder and CEO of Plume Labs. He was out for a run in Paris when he realised that emissions also release pollutants which are an immediate threat to health. “I was focused on the long term environmental risk, when I should have been up in arms about the immediate health impact of pollutants in the air,” he said.
He identified that there was an information gap because of the way air quality is measured. The large air quality units are precise but large, costly and stationary, whereas “air pollution changes from hour to hour and street to street; up to eight times in a city block. So unless you are right next to an air pollution station, it won’t show you the pollution pattern”.
He developed Flow, a personal air quality tracker which can be clipped to a bicycle, pushchair or backpack. Its miniature sensors detect and measure nitrogen dioxide (NO2), particulate matter (PM2.5 and the larger PM10) and volatile organic compounds (VOCs). The measurements are turned into AQI (air quality index) data which is sent via Bluetooth to a connected smartphone. The phone’s GPS pairs the AQI data with a location to render a map of air quality in real-time.
With this data, users can make choices about when and where to exercise for example, which cycle or walking routes to take to minimise exposure to air pollution.
At its launch, 100 volunteers mapped the air quality of 1,000 miles of London’s pavement. “The aim is to build a database to empower decision makers to support air quality or clean air policies, said Lacombe. “Technology alone will not solve climate change or improve air quality overnight but it will make the quality of air much more transparent and if we can empower people to take action to improve then together we can act to bring an end to pollution,” he urged.
