About the performace

[XieMinzhi0624]

Hello! Thank you for your contribution. But there are some problems when I used this code.
I can not reproduce the performance of this code which is shown in the README, for example the env HalfCheetah. I don't know what the problem is. Maybe because I can not download the expert trajectories data in the website in the README. So I used the code to generate expert data. At the same time, I do used the parameter you offered in the code without any change as default. However, the performace is not good (400 average for HalfCheetah whose expert scored 11k+). Could you please do me a favor to give me some suggestion to reproduce the performance? Thanks a lot!