removed "bug" with reward. Attention, currently it is cheaper for an agent to...

removed "bug" with reward. Attention, currently it is cheaper for an agent to wait if we cummulate rewards between the different state!

torch_training/training_navigation.py

@@ -136,7 +136,6 @@ def main(argv):
                 else:
                     update_values = False
                     action = 0
-                    action_prob[action] += 1
                 action_dict.update({a: action})
 
             # Environment step