vuer MuJoCo Environments for Lucid-XR

Setting Up

# place it directly under home.
mkdir ~/fortyfive
cd fortyfive
git clone https://github.com/vuer-ai/lucidxr.git

Now you can go into this repo and install it (and dependencies)

conda create -n lucidxr python=3.11
pip install -e '.[dev]'

The key things to note are: dm_control version needs to be up to date. MuJoCo should be 3.3.4. cloudpickle should be 3.1.1.

Downloading Assets

Need to write this up when we onboard new people.

Scripts:

For detailed notes on how to collect data and how to run experiments, refer to the notes on each of these scripts.

• collect-demo: this one fires up a vuer server, that allows you to collect demonstrations. You can call via

  collect-demo --name pick_block

• visualize-demo: this one visualizes the demonstrations you have collected. You can call via

  visualize-demo --name pick_block

• lucidxr-launch: This one launches on the cluster. You can call via

  lucidxr-launch --sweep some_exp.jsonl 

Training life-cycle

1. run render worker
2. run h5_worker
3. now you can run train. This is a bit too much.