Why the output of integrator remains zero during Reinforcement Learning (RL) training?

I run Deep Q Network to train an agent with the environment being in Simulink. During the training, all outputs of integrator blocks remain zero even though the input is non-zero, as shown below.

If I simulate the same model without RL agent, all is perfectly fine. I work on Matlab 2022a. How to resolve this issue?

Answer 1

[Gunawan]

It is resolved. The reason for the problem is that the learning termination condition is always met at the start of each training step. Therefore, the integration block only outputs the initial condition.