Skip to content

README.md

Latest commit

 

History

History
37 lines (25 loc) · 1.47 KB

README.md

File metadata and controls

37 lines (25 loc) · 1.47 KB
Raw

RoboDojo evaluation

Home · Models · 中文

Download the RoboDojo weights to checkpoints/robodojo.pt, or set ROBODOJO_CHECKPOINT.

Installation

Use an Isaac Sim compatible GPU, CUDA 12.8 toolkit, C++ compiler, and Conda/Miniforge. In the main model environment:

python -m pip install -e '.[modelscope]'
python -m pip install packaging ninja
python -m pip install --no-build-isolation -r eval/robodojo/requirements-policy.txt
export ROBODOJO_POLICY_ENV="$VIRTUAL_ENV"
python -m eval.robodojo.setup --dependencies --assets
bash eval/robodojo/install_simulator.sh
python -m eval.robodojo.download_models

For a Conda model environment, set ROBODOJO_POLICY_ENV="$CONDA_PREFIX". The simulator is installed separately in .venv-robodojo.

ROBODOJO_ROOT selects the benchmark directory (default third_party/RoboDojo); ROBODOJO_SIM_ENV selects the simulator environment. WAM_WAN_PATH and WAM_VLM_PATH override encoder locations.

Run

bash eval/robodojo/evaluate.sh

To evaluate a single task:

bash eval/robodojo/run.sh --tasks stack_bowls

Policy settings are in sim_robodojo.yaml. See RoboDojo for benchmark usage and third-party licenses.