Run agents in habitat simulation enviroment (habitat-lab), and collect data for training.
First install conda environment in docker/:
env.yamlcontains habitat and pytorch lightningenv_habitat.yamlcontains only habitat (more friendly maybe)
Then install habitat-lab locally:
git clone https://github.com/shuhangchen/habitat-lab.git
cd habitat-lab
pip install -e habitat-lab # install habitat_lab
Install utilities packages here.
The current jobs/amlt.yaml is already set up for aml. Note that the docker image at docker.io:shuhangchen/habitat:latest contains both habitat (sim and lab) and pytorch lightning, and it does not depend on the conda environments in docker.