Always predict boundary action in TD3

Hi, I'm having problems with action prediction with TD3. The agent tends to predict boundary actions after it starts learning. Do you know what is causing this problem?

![image](https://github.com/user-attachments/assets/59fbc73b-0ecf-4f0e-bdda-61d98b7c9430)
