Dose this code "Policy Gradient/Doom" really work?

I learn Chapter5 and write Policy Gradient into tf 2.0 according to "Policy Gradient/Doom", and I just wonder if this code is really work. Because after a  night of training, the agent does'nt look like it can recognize aid kit, and the output probability of my Network is just around 0.29~0.35, is my code wrong? 