This guide describes the minimal workflow for evaluating the post-trained OpenPI baseline, including
Install the official OpenPI repository located at: baselines/openpi/third_party/openpi.
Please refer to the official OpenPI README for detailed setup instructions.
i.e.
cd baselines/openpi/third_party/openpi
GIT_LFS_SKIP_SMUDGE=1 uv sync
GIT_LFS_SKIP_SMUDGE=1 uv pip install -e .
# install genmanip client for ebench evaluation
cd path/to/genmanip-client
uv pip install -e .[full_numpy1]
The EBench-specific files are provided under:
baselines/openpi/src/
baselines/openpi/scripts/Layer these files on top of the official OpenPI codebase. This can be done by either updating PYTHONPATH or copying the files to the corresponding locations in the OpenPI repository.
The post-trained OpenPI models on EBench are available at:
Download the desired checkpoint and update the model path in:
scripts/launch_pi_onlineeval.shLaunch the evaluation with:
bash scripts/launch_pi_onlineeval.shPlease make sure that the model path, dataset path, environment settings, and output directory are correctly specified.