meixinzhu / velocity_control Goto Github PK

Dear professor Zhu,

I'm going to work in the field of velocity control using RL and tried to regenerate the results of your paper with the title of "Safe, efficient, and comfortable velocity control based on reinforcement learning for autonomous driving". In the given repository in Github, I could not find the car_following_origin file. Could you please send me this file to be able to generate my own novel results and cite your paper through my publications?

您好，我阅读了您的论文和代码，有如下的一些疑问：

1. 论文中提到TTC阈值取值选择7s，而代码中为何是4.001s？
2. 论文中提到对动作的探索采用的是OU过程，而代码中为何是采用高斯噪声？
3. 论文中提到加速度的范围是[-3,3]，而代码中action_Bound=2，但舒适性奖励时为何仍使用jerk**2/3600来计算？
4. Main文件210-214行有关episode奖励的部分中，为何episode的奖励要用score /= duration这样一种平均奖励来描述而不是直接用完成一个epsode时得到的累计奖励，并且duration = data.shape[0]，其数值等于NGSIM数据集中某个跟车事件的总时刻点数（总时间/0.1），但是若中途发生碰撞，duration的数值应该是小于data.shape[0]的，这个地方不太明白
5. 论文中公式（7）中对于车距的计算公式与simulation_env文件51行的公式不一样，请问有什么考量吗？

Hi,

I really like what you have done in your paper "Safe, Efficient, and Comfortable Velocity Control based on Reinforcement Learning for Autonomous Driving". Will the codes for this paper be released?