SphLA local simulation resources
Resources for the local text-conditioned G1 simulation demonstration. Download
sphla-sim2sim-assets.zip once; the same archive works with both the macOS and
Linux code packages provided on the supplementary website.
This archive contains the deployment EMA model, local text encoder/tokenizer, frozen actor, and robot XML/meshes. It does not contain the training optimizer or training state. The code package is downloaded separately.
| File | Bytes | SHA-256 |
|---|---|---|
sphla-sim2sim-assets.zip |
972,284,074 | 737052365772275061ee11c1876e56d77a381d0dd2243e8024cbcea42c7dfd31 |
Run locally
Extract the code ZIP for your platform and follow its RUN_MAC.md or
RUN_LINUX.md. Keep this asset ZIP intact in a local folder and use the
documented download_assets.py --bundle-dir option to install and verify it.
After installation, the launcher keeps the model and MuJoCo running while a
local browser page accepts successive text instructions and displays execution.
The browser page is a local interface. This repository does not provide hosted inference. Mac execution has been checked in a fresh environment; Linux scripts and dependency availability have been checked, with full Linux execution pending.
Component licenses and provenance
The archive has mixed component licenses. Read the included asset-notices/
files and the code package's THIRD_PARTY_NOTICES.md. The UFO-derived runtime
assets retain their Attribution-NonCommercial 4.0 International terms; the
FLAN-T5 assets retain Apache 2.0 notices. Robot asset provenance and upstream
notices are included. No blanket license or new rights to third-party material
are asserted by this model card.
Use this package for the documented simulation demonstration. It does not provide a validated real-robot deployment or establish task success.