菜单
Run GapONet data preparation. Use when the user invokes /gaponet-data or $gaponet-data, or asks to process, validate, stage, copy, symlink, or prepare GapONet motion .npz data for training or inference.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
GapONet sim-to-real humanoid robot control workflows for Isaac Lab. Use when working on the GapONet repository or tasks involving GapONet setup, environment validation, training, evaluation, checkpoint export, lightweight deployment, DeepONet, Transformer or MLP policy configs, humanoid operator or AMASS tasks, motion data schemas, robot asset integration, or agent-friendly run artifacts.
Run GapONet inference and evaluation. Use when the user invokes /gaponet-inf or $gaponet-inf, or asks to export a GapONet checkpoint to TorchScript/JIT, run lightweight inference, deploy a policy, evaluate test motion data, or produce inference metrics.
Run GapONet training. Use when the user invokes /gaponet-train or $gaponet-train, or asks to train a GapONet humanoid policy after data preparation using Isaac Lab, Isaac Sim, rsl_rl, DeepONet, Transformer, MLP, or the unified runner.
基于 SOC 职业分类
| name | gaponet-data |
| description | Run GapONet data preparation. Use when the user invokes /gaponet-data or $gaponet-data, or asks to process, validate, stage, copy, symlink, or prepare GapONet motion .npz data for training or inference. |
Run the GapONet data stage, not just a review. Work from the GapONet repository root, identified by scripts/prepare_gaponet_data.py, scripts/run_gaponet_job.py, and configs/.
.npz, stage it into the default operator motion path:python scripts/prepare_gaponet_data.py --input-data <input.npz> --output-dir ./runs/data --mode copy --profile operator --force --update-configs
python scripts/prepare_gaponet_data.py --config configs/data_default.json
.npz path. Do not invent or synthesize motion data.runs/data/data_manifest.json status, the staged output_data, and any validation errors.source/sim2real/sim2real/tasks/humanoid_operator/motions/motion_amass/edited_27dof/test.npzconfigs/data_default.jsonruns/data/data_manifest.json--update-configs: configs/deploy_default.json and input_package_example.jsonUse the operator profile by default. It requires:
real_dof_positionsreal_dof_velocitiesreal_dof_positions_cmdreal_dof_torquesjoint_sequencepayloadsUse --profile deploy only when the user explicitly wants lightweight inference validation that does not require the operator training keys.
Consider the data stage complete only when data_manifest.json has "status": "success" and the staged .npz path exists. If Python lacks NumPy, tell the user to activate the GapONet environment before rerunning the data stage.