Intelligent Traffic Signal Control with Deep Q-Learning
A four-way intersection control system combining vehicle perception, traffic-state modeling and a Deep Q-Learning agent in SUMO to reduce waiting time compared with fixed-time control.
My role: M.S. thesis research at Bu-Ali Sina University; perception, state design, reward design, neural-network agent and simulation evaluation.
From vehicle perception to traffic-signal decisions
A four-way intersection control system combining vehicle perception, traffic-state modeling and a Deep Q-Learning agent in SUMO to reduce waiting time compared with fixed-time control.
Four-way intersection with adaptive phase selection
80-cell traffic-state representation and four signal actions
Compared against a fixed-time signal controller across light, medium and heavy traffic scenarios
The perception side later evolved into vehicle-detection publications, while the control side is being extended in a new manuscript
The control side is still evolving.
A new manuscript extending the Deep Q-Learning traffic-control line is currently being prepared for submission. Methodological details are kept concise while the work is unpublished.
Read the thesis note