Ml Agents Brain play different while learning

I Made something similar to Gridword with Grid Sensor.
My Ml Agent learn really nice but when I try play from lasted brain my ML agent play much worse

change request form Fixed update script to Decision Requester but don’t make difference (only play nice while leaning)

Version information:
OS: windows 10
Unity 2020.1.9f1
ml-agents: 0.23.0,
ml-agents-envs: 0.23.0,
Communicator API: 1.3.0,
PyTorch: 1.7.0+cu110

Hey warfanel, your video isn’t playing.
Can you try running mlagents-learn with --inference, --time-scale=1, and --resume? If it plays poorly with that, it’s most likely that the game behaves differently at high time scales. If not, we’ll have to investigate further. Thanks!

I’m sorry video was private.
Now with “–inference, --time-scale=1, and --resume”

is still different while learning
now I have 150+ fps

My project

Scene: “Scene Maze Game”
script: