How to solve the issue where the RL network's output always reaches hmax?

[yumianhuli1]

What variables and training factors influence the final output magnitude? It can't always be fixed at hmax, right?

[zhumingpassional]

pls check the dataset and params