1. Follow install instructions here
Two options:
a) Make a fork (this will be public) to track upstream changes
b) Make a private repo
Create an empty unitialised new repository with your project name, then:
git clone [email protected]:elle-miller/isaaclab_rl_project.git
mv isaaclab_rl_project my_cool_project_name
cd my_cool_project_name
git remote remove origin
git remote add origin [email protected]:yourusername/my_cool_project_name.git
git push -u origin main
# if you want to pull upstream changes
git remote add upstream [email protected]:elle-miller/isaaclab_rl_project.git
git pull upstream main
Log into wandb if not already setup
wandb login
Update the config in tasks/franka/agents/lift.yaml
wandb_kwargs: {
"entity": "%ENTER_YOURS",
"project": "isaac_lab",
"group": "franka_lift",
"name": "test",
}
# the enable_cameras flag is to save progress videos on wandb (can disable via cfg)
python train.py --task Franka_Lift --num_envs 8192 --headless --seed 1234
# play checkpoint with viewer
python play.py --task Franka_Lift --num_envs 256 --checkpoint logs/franka/lift/.../checkpoints/best_agent.pt
In wandb, change the x-axis to global_step. The agent's progress should look like this, but there will be variation between PCs. The learning should also be reproducible for the same seed.
TODO