I have a single-agent RL model in which the dimension of the dimension of the action space is $70$. This action space is too big and the deep RL agent is not learning properly. The boundaries of the action space are $-1$ and $1$.
My question is, how can I reduce the dimensionality of the action space? I have tried to use auto-encoders with random vectors of dimension $70$ between $-1$ and $1$, but it is not working properly. I am training the encoders using a hidden layer with 10 neurons. However, comparing the original action with the result of encoding and decoding it I can see that the average difference between the components is $0.2$ when the action is in the range $[-1,1]$
