| license: other | |
| tags: | |
| - robotics | |
| - imitation-learning | |
| - memory | |
| - imprint | |
| # IMPRINT Datasets | |
| This repository contains the public training stores and simulator environment | |
| assets for IMPRINT. | |
| ## Layout | |
| | Path | Contents | | |
| | --- | --- | | |
| | `simulation/rmbench/450/` | RMBench 450-episode Stage-1 and Stage-2 stores | | |
| | `simulation/rmbench/molmoact2_eval_assets/` | MolmoAct2 metadata and 450 stored language embeddings | | |
| | `simulation/robocasa/600/` | Customized RoboCasa 600-episode shared store | | |
| | `simulation/robocasa/environment/` | Patched RoboCasa evaluation environment | | |
| The stores contain precomputed visual features and causal action/state streams | |
| used by the released training configurations. See `release_manifest.json` for | |
| immutable source revisions, file counts, and byte counts. | |
| Code, configuration, and evaluation instructions are published in the | |
| `Amazon-FAR/imprint` GitHub repository. | |
| Dataset and simulator assets remain subject to their source terms. Public | |
| release does not relicense third-party simulator assets or upstream features. | |