Code for evaluating different visual pre-training strategies.
To install R3M from an existing conda environment, simply run pip install -e . from this directory.
You can alternatively build a fresh conda env from the r3m_base.yaml file here and then install from this directory with pip install -e .
To train policies on top of each representation:
cd evaluation/r3meval/core/
./run.sh
To test transfer with kitchen shift:
cd evaluation/r3meval/core/
./eval.sh
R3M is licensed under the MIT license.
Adapted from the R3M codebase.