BIRMINGHAM, United Kingdom — Could artificial intelligence finally make your morning commute smooth and relatively traffic-free? Researchers from Aston University report that their new AI traffic light system effectively keeps the flow of traffic rolling and mitigates congestion by reading live camera footage and adapting traffic lights on the fly.
Simply put, if there’s no cars coming from the other direction, say goodbye to those long red lights clogging up the street!
The AI utilizes a type of learning called deep reinforcement, which means the program “understands” when it isn’t doing well (traffic is bad) and reacts. As time goes on, the algorithm learns more and more based on better results.
During a round of assessments, this first-of-its-kind AI outperformed all other tested methods. The other methods relied mostly on manually-designed phase transitions.
The research team developed and constructed a cutting-edge, photo-realistic traffic simulator called Traffic 3D to train the AI. Traffic 3D taught the program how to best react to various traffic and weather scenarios.
The AI was then tested on real junction footage. Sure enough, it adapted well to real traffic intersections despite being trained entirely on simulations up until that point. Study authors say this indicates the AI would be effective across many real-world settings.
“We have set this up as a traffic control game. The program gets a ‘reward’ when it gets a car through a junction. Every time a car has to wait or there’s a jam, there’s a negative reward. There’s actually no input from us; we simply control the reward system,” says Dr. Maria Chli, a reader in Computer Science, in a university release.
How does the system differ from regular traffic lights?
Today, most traffic light automation systems at junctions rely on magnetic induction loops, or a wire that sits on the road and recognizes when cars pass over it. The program then reacts to that stimuli. This newly devised AI, however, is able to “see” high traffic volume before cars have even passed the lights. It is much more responsive and can react more quickly.
“The reason we have based this program on learned behaviors is so that it can understand situations it hasn’t explicitly experienced before. We’ve tested this with a physical obstacle that is causing congestion, rather than traffic light phasing, and the system still did well. As long as there is a causal link, the computer will ultimately figure out what that link is. It’s an intensely powerful system,” explains Dr. George Vogiatzis, senior lecturer in Computer Science at Aston University.
Capable of being set up to view any traffic junction, both real and simulated, the AI starts learning autonomously right away. Other areas can be tweaked as well. For example, the reward system can be manipulated to encourage fast passage for emergency vehicles. Importantly, though, the AI always teaches itself – it is never programmed with specific orders.
Ideally, study authors plan on testing the system on real roads this year.
The team presented their findings at the Autonomous Agents and Multi-agent Systems Conference 2022.
