Extract .pt file from the .msh file

[Harry-maximum]

Hi! Recently I trained my own agent in a new robot env, but I find that it uses the mushroom rl so that it save the file format of .msh, so is there any methods I can change the .msh to .pt file?

[robfiras]

Hi! after you load an agent with

from mushroom_rl.core import Agent
path = "your_path"
agent  = Agent.load(path)

You can access the Gaussian policy's mean network with agent.policy._mu.network and save it however you like. Instead of saving and reloading in .msh, you can access and save the mean network similarly during training.

[Harry-maximum]

Dear Firas,

Thank you for your reply! I can now load the .msh file to my a .pt file to the environment. I am working on the GR1T1 and GR1T2 from Fourier Intelligence, the expert trajectory seems very well and behave quite like a human, but after epochs of training, the robot can only walk using small steps. So can you please judge my model and speculate the reasons? The training Hyper_params just keeps similar with Unitree H1 and G1.

Best regards, Xuanbo

run_expert.mp4 https://drive.google.com/file/d/1fZQ9lgWznhHIHovBI6wLWWiKHdieTxVl/view?usp=drive_web

walk_expert.mp4 https://drive.google.com/file/d/1lB7-GKh90Jc-f0qitpwoOX3Fo1pW_XsA/view?usp=drive_web

2024-07-12 09-59-18.mp4 https://drive.google.com/file/d/1D4ehH0u7WVGTSrY-u1hJEFd3LO2dHCR9/view?usp=drive_web

Firas Al-Hafez @.***> 于2024年7月11日周四 18:57写道：

Hi! after you load an agent with

from mushroom_rl.core import Agentpath = "your_path"agent = Agent.load(path)

You can access the Gaussian policy's mean network with "agent.policy._mu.network" and save it however you like.

— Reply to this email directly, view it on GitHub https://github.com/robfiras/loco-mujoco/issues/37#issuecomment-2222628108, or unsubscribe https://github.com/notifications/unsubscribe-auth/A7ND7MLWEZFMB27VQACQJ2TZLZQILAVCNFSM6AAAAABKR7ZUZ2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMRSGYZDQMJQHA . You are receiving this because you authored the thread.Message ID: @.***>