Multi Agent Deep Recurrent Q-Learning for Different Traffic Demands

Multi Agent Deep Recurrent Q-Learning for Different Traffic Demands

Samad, Azlaan Mustafa (TU Delft Electrical Engineering, Mathematics and Computer Science)

Oliehoek, Frans (mentor)
Vuik, Kees (graduation committee)

Delft University of Technology

2020-01-30

In today’s scenario due to rapid urbanisation there has been a shift of population from rural to urban areas especially in developing countries in search of better opportunities. This has lead to unprecedented growth of cities leading to various urbanisation problems. One of the main problems that comes across in urban areas is the increased traffic congestion. This has led to pollution and health issues among people. With the current advancement in Artificial Intelligence, especially in the field of Deep Neural Networks various attempts have been made to apply it in the field of Traffic Light Control. This thesis is an attempt to take forward the problem of solving traffic congestion thereby reducing the total travel time. One of the contributions of this thesis is to study the performance of Deep Recurrent Q-network models in different traffic demands or congestion scenarios. Another contribution of this thesis is to apply different coordination algorithms along with Transfer Learning in Multi-Agent Systems or multiple traffic intersections and study their behaviour. Lastly, the performance of these algorithms are also studied when the number of intersections increase.

Deep Reinforcement Learning
Deep Q-Network
Recurrent Neural Networks

http://resolver.tudelft.nl/uuid:84d20f53-3be7-4e85-8588-b92b962b32fe

Student theses

master thesis

© 2020 Azlaan Mustafa Samad