5789 research outputs found
Sort by
Learning to Control Traffic Lights for Near-Saturated Traffic Flow
No full textTraffic light control method that uses discrete dynamical systems model and feedback control approach is known as both practically and theoretically promising methodology. This is based on describing inflow and outflow at a junction, and deriving feedback gain for green light timing control. One of the key factors is a vehicle’s turning ratio at each junction. However, the value is not directly measurable in real-time, and off-line estimated values are used instead. This paper focuses on the fact that the degree of improvement of traffic flow is measurable by using a specific value is known afterward, and uses Reinforcement Learning method to search for a good turning ratio values through on-line trials
