Artificial intelligence-based traffic lights do not work.  What went wrong?

AI Traffic Lights Not Working. What Went Wrong?

Artificial intelligence in road traffic is no longer well-received by drivers in Hamm. Since the beginning of April, traffic on Ostwennemarstraße has been regulated by new traffic lights. There would be nothing wrong with that, if it were not for the fact that the AI ​​built into it is faulty and often switches the light to red for drivers when there are no pedestrians in sight.

The police have already been called twice in the past three weeks because nervous drivers have had to wait for long red light cycles for no apparent reason, causing huge traffic jams. “Many drivers have called us and complained that they have been waiting for a long time at a red light, even though there are no pedestrians crossing,” says police spokesman Daniel Siegesmund. The AI ​​installed in the traffic lights should recognise such situations and show drivers a green light if there are no pedestrians waiting at the crossing. “However, these signals are not yet as efficient as they should be,” Siegesmund adds.

Image recognition system

The city already has experience with AI traffic lights. The first one was installed in June last year. Cameras connected to it constantly monitor the roadway, bike paths and sidewalks. The image recognition system distinguishes between pedestrians, cyclists and cars and aims to ensure ideal traffic flow. The system recognizes cyclists from a distance of 70 m and calculates their speed, based on which the light changes to green for them at the right time.

The manufacturer praises object recognition as a big bonus of such signals. According to Yunex Traffic, this function also justifies the high price: 70,000 euros, or about twice as much as in the case of classic traffic lights. A company spokesman maintains that the AI ​​signals are able to detect all road users. It focuses on the group most at risk of injury in the event of contact with another vehicle – e-scooter users, among others.

This is also important on Ostwennemarstraße. According to the local traffic commission, before the installation of traffic lights, this street was a black spot on the city map. Road accidents resulting in minor or major injuries occurred frequently. School children were particularly affected.

Pedestrians approaching the crossing have priority. The idea was as follows: built-in cameras recognize when a pedestrian is approaching the crossing. Then the traffic lights switch to red for drivers and green for pedestrians. The more people want to cross the street, the longer the green light cycle for pedestrians lasts (but it never exceeds 24 seconds). When there are no more pedestrians in the range of the cameras, the light changes back to green for drivers.

It is still unclear why the traffic lights sometimes show a steady red light for drivers. According to Hamm city spokesman Tom Herberg, the error probably concerns the interaction of the sensor with the AI ​​software. “The fault is currently being fixed,” he says. However, it is not expected that everything will work perfectly. “As we continue to work on the traffic lights, we will introduce new improvements,” Herberg assures.

Tests in other German cities

While cities like Salzburg and Vienna have been using machine learning in traffic lights for years, such smart lights are only just being tested in Germany. A test project is currently underway in Essenbach, Bavaria. In the town of 11,000 people northeast of Munich, a €100,000 traffic light regulates traffic. There, too, cameras record road users to adjust the light cycle. Among other things, the software recognizes approaching cyclists, records their speed, and adjusts the green light cycle based on this. In addition, an orange flashing light on the traffic light pole warns drivers of potential collisions with cyclists or pedestrians when turning right.

Experts doubt AI traffic light revolution

Experts are skeptical about the trend. One traffic researcher believes that AI installed in traffic lights is more of a PR stunt than a serious technological revolution. Technologies such as radar sensors have been used for years to record traffic data and incorporate it into traffic light control. In some cities, even city buses can control traffic lights using a radio signal.

Others still believe that AI in individual signals brings little benefit. To optimize traffic flow, all the signals in a city would have to communicate with each other. However, the AI ​​models used so far are isolated from the rest of the infrastructure, so the effect is lost.

Google is trying to find a compromise solution with an AI project called “Green Light.” The company analyzed data from Google Maps to allow cities to adjust the green and red light cycles at individual intersections — in this case, using traditional traffic lights. The goal is to make cars flow more efficiently through cities and reduce emissions at intersections by up to 10 percent. Hamburg is one of the cities taking part in the tests.

Despite initial setbacks, the city of Hamm wants to buy more AI-based traffic lights. The first traffic lights installed in June last year also had problems with the red light. In the meantime, however, the AI ​​software has recovered. “So far, everything looks good there. The traffic lights are working fine,” says police spokesman Daniel Siegesmund. Only time will tell whether the new intelligent traffic lights learn quickly enough.

Source: Der Spiegel