-
Notifications
You must be signed in to change notification settings - Fork 13
Description
Hi DreamControl Team,
First, thank you for open sourcing this project. I am currently setting up a native Ubuntu 22.04 environment to deploy these tasks on a Unitree G1.
I have encountered a few blockers while trying to run the Sim2Real inference:
Missing Pre-trained Weights: The README states that "The trained policy weights for most tasks are already in the Sim2Real/logs/ folder." However, I noticed that the Sim2Real/.gitignore explicitly ignores the logs/directory. Consequently, the folder is empty after cloning. Are these weights hosted on an external drive (eg, HuggingFace/Google Drive), or is the intention for users to train from scratch?
Also it would be much appreciated if the weights are provided as my isaac is running on minimum vram possible, thus taking much time.
I am trying the pick and place task
Training Iterations: For the pick and place task, what is the recommended number of iterations to achieve convergence for stable sim-to-real transfer? Is the 4000-iteration mark mentioned in similar contexts the target for this specific task?
Best regards,
Zubair