Does MolmoAct2 run on I2RT YAM?
Capabilities
| Capability | State | Notes | Evidence |
|---|---|---|---|
| Train / fine-tune | works | Fine-tuned on the 720-hour MolmoAct2-BimanualYAM dataset and per-task on bimanual YAM data. | ev-molmoact2-paper-yam, ev-molmoact2-bimanualyam-card |
| Real-robot inference | works | Bimanual YAM setup (two arms, top/left/right cameras). LeRobot v0.6.1 also documents a LeRobot-format checkpoint for lerobot-rollout, but LeRobot has no built-in YAM robot type, so that path needs a third-party YAM plugin. | ev-molmoact2-paper-yam-eval, ev-molmoact2-yam-example, ev-lerobot-molmoact2-doc |
| Simulation inference | no verified evidence | — | |
| Teleop + record | no verified evidence | — |
Evidence 5 sources; the cell shows the highest tier
- T1 · upstream documentedpapersource 2026-05-04verified 2026-10-24 (1d ago)
MolmoAct2 paper: the MolmoAct2-BimanualYAM dataset was collected on the authors' bimanual YAM setup and used to train MolmoAct2.
“All data is collected on our custom bimanual YAM (Yet Another Manipulator) setup, shown in Figure 3.”
https://arxiv.org/abs/2605.02881 - T1 · upstream documentedpapersource 2026-05-04verified 2026-10-24 (1d ago)
MolmoAct2 paper: real-world fine-tuning evaluation on bimanual and mobile bimanual YAM, 8 tasks x 50 trials (average success 50.1%).
“For real-world evaluation, we conduct a large-scale, systematic study on both a bimanual YAM setup and a mobile bimanual YAM setup, covering 8 tasks”
https://arxiv.org/abs/2605.02881 - T1 · upstream documentedmodel-cardverified 2026-10-24 (1d ago)
Official allenai/MolmoAct2-BimanualYAM checkpoint, fine-tuned for bimanual YAM inference (absolute joint-pose control).
“It is intended for both further fine-tuning and bimanual YAM policy inference.”
https://huggingface.co/allenai/MolmoAct2-BimanualYAM - T1 · upstream documentedsource-codeverified 2026-10-24 (1d ago)
Upstream molmoact2 repo ships a closed-loop bimanual YAM eval client that drives real YAM arms through the I2RT driver.
“Hardware-coupled example: it talks to real YAM arms over CAN (via `i2rt`) and”
https://github.com/allenai/molmoact2/blob/6070080a20321b4f498ab30f28e1d09ac465edb7/examples/yam/README.md - T1 · upstream documentedofficial-docssource 2026-08-03verified 2026-10-24 (1d ago)lerobot 0.6.1
LeRobot v0.6.1 MolmoAct2 doc lists lerobot/MolmoAct2-BimanualYAM-LeRobot under Hardware Deployment (lerobot-rollout) with YAM camera keys.
“MolmoAct2-BimanualYAM-LeRobot | `top`, `left`, `right` | YAM 3-camera setup”
https://github.com/huggingface/lerobot/blob/v0.6.1/docs/source/molmoact2.mdx
MolmoAct2 inference speed by hardware robot-independent · inference cost only
| Compute | Runtime | Precision | Cams | p50 / p95 ms | Inference | Mem | Platform | Source |
|---|---|---|---|---|---|---|---|---|
| H100 80GB | PyTorch eager | bf16 | 2 | — | — | 13 GB | lerobot 0.6.1conditionsos: Ubuntu 22.04 (repo tested with) · memory_sampling: nvidia-smi peak Inference memory only (12.1 GiB peak GPU memory = 13.0 GB); no latency reported. LIBERO with two RGB cameras, chunk_size=10, 8 flow timesteps. Not task success. | T3 · community reported |
Embed this cell
Badge for READMEs and model cards. It always shows the current state.
[](https://caniembody.pages.dev/compat/i2rt-yam/molmoact2)<iframe src="https://caniembody.pages.dev/embed/i2rt-yam/molmoact2" width="380" height="132" style="border:0" title="MolmoAct2 on I2RT YAM | CanIEmbody"></iframe>History
- 2026-10-25 fecccd1 Initial release: robot policy compatibility matrix · Danilo Znamerovszkij