Services
此内容尚不支持你的语言。
The model processes skills and model roles call; a deploy says where each runs (Services and deploys). VRAM is measured on a 24 GB RTX 3090.
| Service | Kind | What it is | VRAM |
|---|---|---|---|
| dm05 | policy | DM0.5 VLA with a memory of past frames (Dexmal DM05-MEM-Robodojo-Sim), bf16, through XPolicyLab’s OpenDM adapter | 13 GB |
| g05 | policy | G0.5 VLA, RoboDojo’s own checkpoint (joint actions, both arms), through XPolicyLab’s G05 adapter | 16.4 GB |
| kinematics | kinematics | ARX X5 kinematics (fk, ik, plan_path) through RoboDojo’s cuRobo planner, without the simulator | 2 GB |
| mcp-units | mcp | An example MCP tool server (unit conversion) for the mcp skill | – |
| molmo2 | pointer | Molmo2-8B open-vocabulary pointing (point_to, localize, localize_3d), int8 with the vision tower in bf16 | 12.7 GB |
| openwam-policy | policy | OpenWAM-alpha world-action model run as a policy (end-effector actions), RoboDojo’s own checkpoint, through XPolicyLab’s OpenWAM adapter | 15.3 GB |
| openwam | world_model | OpenWAM world-action model trained on RoboDojo (predict_future): predicted camera views and end-effector motion | 15.3 GB |
| pi05 | policy | pi0.5 VLA, RoboDojo’s own checkpoint, through XPolicyLab’s Pi_05 adapter (openpi, JAX) | 16 GB |
| qwen3-vl-32b | llm | Qwen3-VL-32B-Instruct in 4 bit, self-hosted: the development model that writes programs (and answers VLM questions) | 21 GB |
| qwen3-vl-8b | vlm | Qwen3-VL-8B-Instruct, self-hosted: questions about images and pointing for development runs without API spend | 18 GB |
| sam3 | segmenter | SAM 3 instance segmentation from text or point prompts (segment_text, segment_point; inside pick and place) | 7.2 GB |
| simate | policy | Simate-beta VLA with a memory of past frames, RoboDojo’s own checkpoint, through XPolicyLab’s Simate_beta adapter | 14 GB |