|
چکیده
|
In the past decades, urbanization, growth in urban population, and the increasing use of private vehicles have created significant problems in the transport systems of urban areas, such as traffic congestion, rising commute times, fuel consumption, and lowered levels of safety. Urban intersections or junctions represent important critical spots in traffic and influence the efficiency of transport systems significantly. The regulation of traffic lights at intersections affects traffic efficiency and urban safety directly.
The traditional control systems of traffic signals, namely Fixed Time Control and Actuated Control, although easy to implement and reliable, are highly dependent on historical information and cannot respond immediately to dynamic changes in traffic flow. Consequently, these systems cannot adapt to changes or fluctuations caused by traffic demand, unexpected incidents, or route changes of travelers, since they are not capable of reacting to uncertain and dynamic situations immediately or in real-time situations.
However, with the groundbreaking innovation in the area of artificial intelligence (AI) and machine learning technology, especially in deep reinforcement learning (DRL), novel techniques have been developed in the area of intelligent traffic signal control. These techniques are able to represent traffic control as a sequential decision problem so as to be able to learn the optimal control policy against the environment. However, safety concerns are usually not considered in the process of the technique utilized in most research.
If safety considerations are overlooked within DRL-based systems, the consequences may include dangerous decisions like sudden phase transitions, illegal reductions of safe times, and more conflict points between cars and pedestrians. Because the task of traffic systems inherently involves safety, a wrong choice may have irreversible impacts. As such, the forefront issue that the research aims to solve is the developm
|