Reinforcement Learning based quadrotor control using Soft Actor-Critic (the reward is not converging)

Hi, I am trying to control of a rotary wing UAV (quadrotor) by using Soft-Actor Critic methodology, but I have some problems, my reward is increasing continously after the point you see following image, what is the main problem, can you advice for this situation, I am sharing my files (Simulink and m-file). My max reward values should be zero as we define in reward function on Simulink file. This reward function indicates that the difference between desired trajectory and actual trajectory is about zero.

[SALMAN IJAZ]

Hello, can you share your simualtions for detailed understanding and discussion.