Reinforcement Learning trained Agent has same output every time

I trained an TD3-Agend and the results were verry good, I watched the scope during the training so I know it does what it should, but unfortunately the simulation does literarlly nothing.

The Output of the Agend doesnt change, its just the same every time, unless the observations change...

Does anyone know how this can happen? Or what I can do diffrent?

[Emmanouil Tzorakoleftherakis]

I am not sure how to interpret "The Output of the Agend doesnt change, its just the same every time, unless the observations change". If the observation input is the same, the output will also be the same with TD3 since it's deterministic. Can you clarify?

[Jannick Rowedder]

Thanks for your reply!

I would go with it, if the input is the same, but it isn't. I run the simulation with a local reset function so the input is diffrent every time. I can see it in the tracked values from simulink and doesn't matter what inputs there are, the output stays the same aufter training.

I had a closer look on the training as well and there the agend changes the output and the reward increases so basically everythink as it should, but aufter training the agend is useless.

I want to tune a cotroller by a TD3-Agend, so there is just one action, based on the observations right in the beginning.