Reinforcement learning agent does not perform any action in simulation after train

Data store memory blocks were added to prevent algebraic loop errors:

While training process, the scope shows active action signal (yellow):

But after training, in simulation, or running simulink model, the action or rl agent stay idle. Non zero, but constant:

What can be the cause?

How can i enable actions, when run simulink model?

[Alexey J]

It turns out that training at 2400 episode, converges precisely to such a result..

Answer 1

[Alexey J]

Solved with multiplication x2 reward and x10 action signal.

There is no hidden settings.

The cause is somewhere in reward algorithm.

Next time i advice to play with action and reward signals, to fill all provided space, while training.

[Anh Tran]

I can totally relate with you Alexey. Reward shaping plays a big role in the design of an RL agent. It would be great if you could share what reward function did you come up with to solve the task and how did you iterate to get a better function.