Traffic Flow Test Platform

Traffic flow test platform, especially for reinforcement learning, named TFTP.

A traffic signal control framework that can combine a variety of reinforcement learning algorithms, traditional algorithms and discrete reinforcement learning algorithms. It has two environments: cityflow and sumo. Several algorithms have been implemented.

How to run

check and modify the main.py in the root folder, then run it.

How to use

this is still an on going project, some algorithm might be not implement or occur some error. but the idea is simple, you can use it to test your idea, add new engine except for sumo and cityflow, add your specific map mode, develop new state, action, reward information, and so on.

this project is NOT implement fully, but can be used now

folder describe

folder describtion
algs algorithm
data traffic files and information
configs config class
misc tools maybe used by the project
envs environment including sumo and cityflow
records the output on the running process
tmp for debuging, tesing the gramma, and so on.

file describe

file name information
main.py main entrance

Structure of round learner

round_learner

the UML of the class

uml_class

效果

sumo

ecust_compus_sumo

cityflow

ecust_compus_cityflow

Curve example in dqn

dqn_vehicle

dqn_reward

Note

本代码框架部分思路来源于,感谢他们在算法和程序上做的贡献。

https://github.com/gjzheng93/frap-pub

https://github.com/zxsRambo/metalight

CityFlow

https://github.com/cityflow-project/CityFlow

Sumo

https://www.eclipse.org/sumo/

Contract

Mr. Bai:[email protected]

GitHub

View Github