2022/6/13

Posted 2022-06-15 yellowyyyy

The Reforce Leaning based on Q-learning method, which is used in the interactive control of autos in the one single intersection. Easily speaking, the RL is one kinds of artificial intelligence, without man supervise, and after a vast number of exercise or trains while we give computer a goal and the correspondingly reward, it will do the things or the ways most efficient——find the something we what.

So, I make an interesting experiment with Sumo (a traffic simulation software) and the RL which play in python. First, we can use the TRACI to connect Sumo and python, after this, I can do anything I want in Sumo by python. The specific programming steps will not be introduced here, and the experiment’s result will show as follow.

The picture1 shows the reaction of a man who drive to the intersection with a relatively higher speed. And the picture2 shows the decision of RL agents after trains, obviously, it make speed change and through the cross more steadily——which do not stop at the line.

用基于Q-learning的强化学习对在单点交叉口进行交互式控制。简单地说，强化学习是人工智能的一种无监督学习方式，我们给电脑一个目标，并给予相应的奖励，经过大量的训练后，它就会变得高效——比如很快找到一个东西。
因此，我用Sumo（一种交通仿真软件）和RL（在python中完成）进行了一个有趣的实验。首先，TRACI是一种连接Sumo和python的工具，有了它，我就可以通过python在Sumo中做很多事情。这里不介绍具体的编程步骤，实验结果如下。 图1显示了人类的驾驶，以相对较高的速度驶向十字路口，然后停车等待。而图2显示了RL智能体在经过训练后的决定，显然，它通过速度变化使车辆更加稳定的通过交叉口——也就是不停车。