LST

Experimental robot-policy checkpoints. Upload is in progress; absence of a per-family VERIFIED manifest means that family is not yet a complete verified archive.

Original run names distinguish experiments; directory names alone are not proof of their effective settings or evaluation results. Inspect the associated configuration and normalization files. These custom checkpoints are not generic plug-and-play models and must not be used on hardware without the matching software and safety checks.

Initial collection: RDT-1B

Fine-tuned robot-policy checkpoints derived from RDT-1B. Upstream RDT code and pretrained weights are MIT licensed; see licenses/RDT-MIT.txt. Files in this collection are modified/fine-tuned derivatives, not the upstream release. Other model families are not covered by the RDT license and are added only after review.

Files are preserved byte-for-byte. PyTorch checkpoints can contain training state and require the corresponding custom architecture. Prefer safe loading; never unpickle untrusted model files. Original datasets, videos, credentials and logs are not included.

An archive upload is not a claim that an experiment passed evaluation. No local source file is removed automatically by this process.