| license: apache-2.0 | |
| language: | |
| - en | |
| - zh | |
| tags: | |
| - robotics | |
| - world-action-model | |
| - robodojo | |
| - safetensors | |
| # WorldScape-R2SI | |
| WorldScape Policy 2 checkpoints post-trained for RoboDojo. | |
| Both variants support `arx_x5` absolute joint control in interactive mode. | |
| ## Models | |
| | Model | Description | | |
| | --- | --- | | |
| | [Memory](wsp_2_memory_posttrain_robodojo/README.md) | Rolling context of up to 24 chunks | | |
| | [Flash](wsp_2_flash_posttrain_robodojo/README.md) | Single-chunk context | | |
| ## Usage | |
| Download the repository, then point your downstream policy to the desired model directory: | |
| ```bash | |
| hf download manifoldai-research/WorldScape-R2SI --local-dir ./WorldScape-R2SI | |
| ``` | |
| Each model directory includes its weights, tokenizer, configuration and normalization | |
| statistics. Use the settings supplied with that model. Installation and inference | |
| are handled by the downstream policy. | |
| ## License | |
| [Apache-2.0](LICENSE). Attribution and release details are included in each model directory. | |