DESIGN AND IMPLEMENTATION OF FOUR LANE INTERSECTION MODEL FOR Q-LEARNING ACCELERATOR TESTING VISUALIZATION IN CONTROLLING TRAFFIC LIGHT

• Main Author: Widyanto, Dismas
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/66525
• Institution: Institut Teknologi Bandung

ii ABSTRACT DESIGN AND IMPLEMENTATION OF FOUR LANE INTERSECTION MODEL FOR Q-LEARNING ACCELERATOR TESTING VISUALIZATION IN CONTROLLING TRAFFIC LIGHT By Dismas Widyanto NIM: 13218065 (Bachelor????s Program in Electrical Engineering) Congestion is a common problem, especially in big cities. One of the causes of congestion is waiting times at intersections with traffic lights. The current system controls traffic lights alternately at fixed time intervals. Such a system cannot adapt to the junction conditions at that time. Control using a reinforcement learning algorithm can be used to develop an adaptive system specifically using q-learning. Q-learning is able to provide a response based on the current conditions of an environment. Research on Qlearning has been widely conducted. Some of them developed an accelerator architecture to speed up the calculation process. The use of q-learning in the transportation sector is also starting to be widespread. However, the platform for testing q-learning accelerators in controlling traffic lights is still rare. Therefore, we need a testing platform that can test the performance of the q-learning accelerator in controlling traffic lights. The development of this platform aims to evaluate the performance of the q-learning accelerator. The design process consists of problem formulation, specification preparation, design process, implementation, and specification verification. The results of the test platform creation show that the product can be used to demonstrate the performance of the q-learning accelerator. Tests show that RL control can provide a 3% reduction in vehicle queues compared to conventional control. Keywords: accelerator, design, q-learning, traffic.