INTELLIGENT TRAFFIC LIGHT CYCLE CONTROL USING MACHINE LEARNING

Abstract

When there is a red signal for a longer duration vehicles takes longer time to pass the road which
may cause traffic jam and also increases the queue length. To address this issue, we have to operate
on the existing infrastructure by improving the systems that have a significant impact on the traffic
flow, such as traffic signal controllers. This approach is desirable since it has a low cost of
implementation. Therefore, the recent advancements in the field of Machine learning allows us to
propose a Deep Reinforcement Learning algorithm in which the learning agent is able to control a
traffic light system, which aims to increase the efficiency of road transportation. The agent makes use
of a deep neural network to choose which light phase activate and it is trained using the Q-learning.
The traffic interaction is handled by the traffic micro-simulator SUMO that is used to let the agent
experience a multitude of situations and learn by mistake which effects have the light phase activated
during the simulations.