Open MullerLin opened 2 years ago
Hi @MullerLin,
glad you enjoy the project! The clipping/normalization of abs and actions obviously is intended to help the stock RL agents in stable-baselines3 to more easily learn. There isn't any specific reason for the plots to follow the same clipping/normalization other than the simplicity of using the data as they come out of the aviary environment class.
Hi @MullerLin,
glad you enjoy the project! The clipping/normalization of abs and actions obviously is intended to help the stock RL agents in stable-baselines3 to more easily learn. There isn't any specific reason for the plots to follow the same clipping/normalization other than the simplicity of using the data as they come out of the aviary environment class.
Thanks a lot for your reply! It helps me to understand the project better!
Hi, thanks to your wonderful project, I am working on training some RL tasks on the project, and I use test_multiagent.py/test_singleagent.py to see the training results, like how the drone act. I find that the final plotted data curves from logger are clipped and normalized because the logger logs the data from obs , the data after clipping and normalizing. It made me confused at first when I tried to trained the drone to fly some certain position, and I modified data on the curve by multiplying the normalization coefficient.
I am wondering whether there are some reasons to show the clipped and normalized data on final curve in the two test.py, like helping analyze or some other reasons.
Thanks a lot.