Description issues

[Mvp-Evan]

From the code, it is more like a DDPG algorithms but not PPO algorithms. Besides, for the auto-encoder part, this code is not use auto-encoder to extract features of input image, which looks use encoder as the CNNs. Why not learn and update the weight of these CNNs with actor and critics network? Why need to train these CNNs separately?