An artificially intelligent traffic light system is being developed and tested by a team at Aston University in Birmingham, UK. The team aims for this technology to assist in maintaining traffic flow and ease congestion.

According to an article in The Engineer, the AI can begin to autonomously learn from any junction it is set up at, whether real-world or simulated, using a technique called Deep Reinforcement Learning. It effectively understands when it is not performing well and adapts. It can also be guided towards teaching itself to allow emergency vehicles through when needed.

An important step has been the use of Traffic 3D, a realistic traffic simulator programmed with various traffic and weather scenarios. This tool, also created by the researchers, has been used alongside a rewards system to develop the adaptation skills of the AI. 

The average driver is estimated to waste 115 hours and £894 worth of fuel in a year, largely due to problems with traffic light timings., The AI system these researchers have created is, already outperforming other methods of traffic light control, and could result in large efficiency gains.