Skip to content

Training smart traffic signals through reinforcement learning using Unity3D and MLAgents

License

Notifications You must be signed in to change notification settings

ethicia/AITMS-Sim

Repository files navigation

AITMS-Sim

Summary

Traffic congestion is becoming one of the critical issues in cities with increasing population and number of vehicles. They not only cause problems like delays and stress to drivers but also cause secondary problems like increasing fuel consumption, transportation costs and pollution.

The causes of congestion can be divided into two categories, recurring and non recurring congestion. Recurring congestion can be expected to occur at the same time every weekday as a result of high volumes of commuter traffic traveling on roadways that are at or near their carrying capacity. Non-recurring congestion occurs as a result of an unexpected or non-typical event. Some causes of non-recurring congestion include: vehicular crashes, vehicle breakdowns, roadway construction, inclimate weather, and additional traffic resulting from special events. While non-recurring congestion can be unpredictable and difficult to treat, recurring congestion can be reduced by increasing road capacity or with the help of adaptive traffic control systems.

There are several existing standardized solutions for adaptive traffic control such as SCOOT, SCAT, etc. which have been implemented in many major metropolitan cities. However, most suburban and urban areas use conventional traffic control systems such as manual traffic control or non adaptive automated traffic control. Manual control consists of an on-site traffic official guiding vehicles. Non adaptive automated traffic control refers to the use of fixed timers in traffic signals. Wide implementation of standardized adaptive traffic control is not possible due to lack of feasibility since it requires manual labor and installation of new sensors. Therefore a more feasible solution which reuses existing infrastructure is required.

This project proposes an Artificially Intelligent Traffic Management System which uses existing CCTV feed to optimize traffic control over small to medium scale road networks. It uses an artificially intelligent agent or model to handle the complexity of day to day traffic in real-time. Furthermore, simulations will be performed to demonstrate and test the model.

Report

Click here to view project report.

About

Training smart traffic signals through reinforcement learning using Unity3D and MLAgents

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •