Doubts about the code

[Hznnnnnnn]

Hello, after reading your paper, when I reproduced your code, I was not sure whether the model in "risk-sensitive-rl-master/art-iqn/experiments/sim/IQN.pth" was the model you trained and used in the paper. When I ran tactical.py with this model, I found that the navigation would only go straight forward and would not turn to achieve the effect in the paper. What is the reason?

At the same time, when I train myself, the success rate reaches about 0.88 after 4000-5000 episodes of training, but it is rarely successful when running tactical.py. Is there some problem in tactical.py?