21 Dec 2020 News

NoTraffic’s Smart tech is revolutionizing the traffic light

Every motorist has experienced the frustration of waiting at a red light when there is no traffic in sight. These days, countless household devices use sensors, cameras and connectivity to understand and respond to the world – so why do most traffic lights languish in the analog age, causing our vehicles to unnecessarily stop and idle?

In a spring 2019 report, Guidehouse Insights forecast that just 7% of North America’s signalized intersections would use so-called adaptive traffic-control technology in 2020. On a global level, that number was 2.3%.

“Ninety-nine percent of traffic in the U.S. is not reacting to what’s happening on the road,” said Tal Kreisler, the CEO of NoTraffic, based in Tel Aviv, Israel and Palo Alto, Calif. The company, founded in 2017, is among a field of startups and legacy players trying to accelerate the shift from analog traffic-light controls to digital, adaptive systems using artificial intelligence (AI). Another firm, Pittsburgh-based Rapid Flow Technologies, began commercializing its signal-control technology dubbed Surtrac not long after it was spun out of the Robotics Institute at Carnegie Mellon University in 2016.

Centralized vs. Distributed strategies

“We are taking a completely decentralized approach,” said Griffin Schultz, CEO at Rapid Flow Technologies. The Surtrac (Scalable Urban Traffic Control) software is designed to optimize traffic-control signaling to reduce congestion on an intersection-by-intersection basis. The scalable approach can start at a single intersection, growing to a city’s entire traffic network. “Once the configuration files are loaded, it runs without human involvement,” said Schultz.

By contrast, NoTraffic offers an end-to-end solution that includes software, a camera-radar sensor unit and a SAS-based traffic-management platform. An initial deployment for NoTraffic usually comprises about 10 to 15 intersections, although some setups are larger.

In April 2020, Florida’s Broward County completed a comprehensive study of 15 adaptive traffic-control systems. About half are distributed systems that process data at each local controller and then communicate directly with the adjacent local controllers. The others are centralized systems that process the optimized signal system parameters with a central server and then push those parameters to local controllers. Most are compatible with the same vehicle-detection hardware, such as inductive loop (embedded in asphalt), video and radar.

What primarily distinguishes one adaptive system from the other is the unique algorithm. The formulas differ based on their combination of traffic-theory timing strategies: cycle length, split time, offset time and phase sequence.

Smarter, responsive intersections are shown to reduce congestion and emissions by 20% or more. Forty-five percent of U.S. mayors say traffic is their city’s top problem, according to a 2018 study by Bloomberg Philanthropies. Nonetheless, cash-strapped cities have few incentives to shoulder the cost of relatively expensive new technology. And cities often hesitate to trust their traffic to startups with technology not yet proven at scale. Meantime, traffic-management industry giants such as Siemens and Econolite continue to evolve their offerings. The latest version of the Siemens-backed SCOOT (Split Cycle Offset Optimization Technique) system uses real-time optimization. Guidehouse Insights estimates that some version of SCOOT has been deployed at about 45,000 intersections globally.

Detect, control and communicate

Schultz, the Rapid Flow CEO, described the company’s solution as an AI robotic system. There are three primary components. First is a detector — essentially a set of sensors. Surtrac is agnostic about what provides those sensor inputs. Data can come from video cameras, radar (useful in rain and snow) and inductive loops. Eventually, the inputs could come from cell phones, connected vehicle systems or even navigation routes voluntarily shared by drivers.

“We can integrate with almost anything that’s there,” he said. Surtrac relies on third-party sources for the object classification — the software that identifies an object as a car, firetruck, bicyclist, etc. After receiving a feed about the number and type of vehicles and their heading, Surtrac generates an optimization plan. That plan models the location of all those traffic objects from one second to as long as five minutes.

The second component, a command controller, then kicks in. The Surtrac software runs on a commodity computer in the intersection’s control cabinet. It instructs the light to change according to the optimization plan. However, the system’s real power comes with the third component: communication from one intersection to the next. Once again, Surtrac relies on existing infrastructure, ranging from fiber optics, WiFi, and cellular communication. The data also is shared to the cloud — primarily so it can be monitored.

Improved traffic flow can occur at a single intersection. But Schultz said, “We can’t deliver the amount of value that we want to deliver until we’re at nearly every intersection in the city.” One of Surtrac’s biggest implementations, mostly in place since about 2016, is at 50 intersections in Pittsburgh, which has some 600 signalized intersections.

Schultz declined to pinpoint the per-intersection cost to install Surtrac. But he said that the oft-quoted price of $20,000 per intersection for a conventional four-camera analog system was a “reasonable assumption” to make for Surtrac’s fee. The company doesn’t replace existing infrastructure; the fees are additive. Ryan Citron, senior research analyst at Guidehouse Insights, pegged the per-intersection cost for an adaptive system at $30,000 to $60,000.

Alternative plug-and-play approach

Competing NoTraffic offers an all-in-one, hardware-software solution. The company’s technology also includes three essential components. First is a set of four proprietary sensors –one per approach – that uses a camera with machine vision and a compact automotive-grade radar in a box. The system can take other inputs as well, including signals from connected vehicles.

The sensor units detect objects approaching the intersection from up to 900 feet (274 m) away. The system differentiates traffic based on a human-eye level, estimating each vehicle’s arrival time to the intersection. NoTraffic situates an Nvidia-powered “edge-computing device” in the intersection cabinet, although some of its AI is shared with cloud-based computers. “At any given second, our edge communication device is running a very complex algorithm that calculates all the possible scenarios like a chess game,” said Kreisler. “It works not only locally on one intersection. It knows what’s happening in the next intersection through the cloud.” The edge-computing device also acts as the controller.

Kreisler estimates that an installation takes about one hour. “We simply have to connect to a power source and it’s up and running,” he said. NoTraffic has agreements with companies such as AT&T to perform installations.

Finally, NoTraffic uses cellular-based communications to create a network effect via the cloud. Rather than passing full-streaming video or other large files, a set of extrapolated lightweight metadata data about the vehicle types, latitude-longitude, speed and direction is passed between the nodes. Latency is claimed to be minimal.

Click here to read the full article by Bradley Berman, originally published in sae.org