Stationary action values in reinforcement learning

Hello,

I'am trying to train reinforcement learning with high dimensional actions.

ex)

ActionInfo = rlNumericSpec([9 1], ...

'LowerLimit',[0.012 0.01 0.015 0.05 92 3 0.1 3 0.012]',...

'UpperLimit',[0.030 0.03 0.1 0.3 100 5 0.3 5 0.040]');

and construct my actornetwork with "scalingLayer" and low value of neurons

ex)

but only 3~5 parameter changes (other parameter doesn't change)

ex)

Is there any solution to make all action parameters changeable???

Thank you for help.

[Venu]

Hi @minhyuk jeung,

Can you provide some information regarding your learning environment, ObservationInfo?

Thanks

Answer 1

[minhyuk jeung]

Thank you for your help.

Fortunately, this problem was solved with changing the agent from DDPG to SAC.