I have a model with TD3 + lstm in both actor and critic.
I am trying to make it learn to predict some specific actions based on the environment conditions. However i see that the AI predicts very nearly the same actions for all the states.Moreover my problem statement has many negative rewards and there is only one kind of scenario is which the the reward is positive. In this scenario the AI needs to get the state and past state and a particular action gives high rewards. rest all gives small negative reward.
Also the action it is giving is not the optimal negative loss avoidance action
Can someone please help to let know the intuition what is wrong happening here.