RiemannRSI models and resources
This repository contains the Riemann1.0 learned policy and the exact public vision models and small task detectors used by its RoboDojo integration. Mirrors preserve model parameters and original upstream identities; they are not additional Riemann-trained foundation models.
| Directory | Contents |
|---|---|
| Riemann1.0 | Learned policy, normalization and inference-only configuration. |
| common_visual_models | SAM3, CLIP, GroundingDINO and the Wan2.2 encoder/VAE/tokenizer used at inference. |
| task_detectors | The conveyor and imitation-sorting vision classifiers required by the original rule implementations. |
Use the fixed revision and content hashes declared by the RiemannRSI source release. Do not replace a pinned model with a similarly named checkpoint. No new task success rate or official-cloud acceptance is claimed by publishing these artifacts.
resources/: deduplicated task templates and public static reference data.
Licenses
Licenses apply per component. Riemann1.0 retains its declared Apache-2.0 model-card license; SAM3 retains the Meta SAM License; CLIP retains MIT; GroundingDINO and Wan2.2 retain their upstream notices. License copies and source identities are included alongside each mirrored public model. The root metadata does not relicense third-party models. Task detector provenance is documented in its directory; no additional ownership or training-data claim is made.
- Downloads last month
- -