How to extract "Actor" part of a pre-trained agent (RL - DDPG)

1 visualizzazione (ultimi 30 giorni)
I have trained a DDPG agent and it works fine. I want to extract the "Actor" network of this pre-trained agent. I don't need the critic network after the training process. The "Actor" is my controller. The attached picture is my system and agent. Disconnecting the two signals "isdone" and "reward" has no impact on the result after training the agent. I want only the "Actor" part so that I use it for linearization. If I connect these two signals, the linearizations fails. I have already used the "predict" block with a saved mat file "actorNet", no result is obtained.
Thanks.

Risposte (0)

Prodotti


Release

R2021a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by