MS-HAB Runtime

MS-HAB evaluation normally runs outside the caller's Python environment. praxis-eval starts a policy endpoint and launches the MS-HAB runtime as a subprocess.

Command

praxis-eval-setup mshab

Useful options:

praxis-eval-setup mshab --env-manager micromamba --env-name mshab-praxis
praxis-eval-setup mshab --ms-asset-dir /data/mshab/maniskill_assets
praxis-eval-setup mshab --skip-assets
praxis-eval-setup mshab --skip-status-check
praxis-eval-setup mshab --help

What It Does

The setup helper:

1. Locates the conda environment spec shipped by praxis-mshab: mshab/praxis_conda_env.yaml.
2. Creates or updates the requested conda or micromamba environment.
3. Installs praxis-mshab, praxis-remote, and praxis-eval into that environment.
4. Downloads ManiSkill assets: ycb, ReplicaCAD, and ReplicaCADRearrange unless --skip-assets is set.
5. Installs activation hooks for Vulkan/EGL and MS_ASSET_DIR.
6. Runs an import/status check unless --skip-status-check is set.

Default managed asset root:

~/.cache/praxis_eval/assets/mshab/maniskill_assets

Evaluation Config

Pass the runtime interpreter to evaluation:

from praxis_eval import EvalConfig

config = EvalConfig(
    task="set_table",
    num_eval_per_task=3,
    num_parallel_env=1,
    output_dir="eval/mshab_set_table",
    env_kwargs={
        "python_bin": "/path/to/mshab-praxis/bin/python",
        "ms_asset_dir": "/data/mshab/maniskill_assets",
        "split": "val",
        "obs_mode": "rgb",
    },
)

Verify

praxis-eval-verify mshab --env-name mshab-praxis --num-episodes 1 --num-envs 1
praxis-eval-verify mshab --env-python-bin /path/to/mshab-praxis/bin/python

Caveats

• The simulator runtime must be able to import mshab, praxis_eval, and praxis_remote.
• The runtime expects ReplicaCAD rearrange task plans and spawn data below MS_ASSET_DIR.
• GPU cluster nodes often need explicit Vulkan/EGL variables; setup writes activation hooks for that reason.