Analyzing the Impact of AI-Driven Traffic Signal Control on Reducing Congestion and Improving Road Safety

Abstract

A considerable proportion of road traffic congestion is generated by the inefficient control of traffic signals at road intersections. We propose an artificial intelligence-driven, adaptive traffic signal control approach to reduce traffic congestion,as well as improve road safety by accommodating the unsignalized  movement of vulnerable road users. Specifically, we adopt a curriculum learning mechanism, in which the overall AI-driven traffic signal control problem is separated into multiple sub-optimized problems dealing with different scenarios trapped in the training data, and the learned results are gradually sampled and added to the training data for global problem optimization. We evaluate our work through large-scale simulations, and the results demonstrate the clear and consistent improvement of AI-driven traffic signal control in reducing congestion, where the longer the traffic volume peak, the more significant this improvement can be observed. More interestingly, our approach dramatically reduces road conflicts between traffic participants across different intersection scenarios and hence improves road safety as well.