MolmoBot-SPOC-RBY1Rigid

[Paper] [Project Website] [Code] [Data]

MolmoBot-SPOC-RBY1Rigid is the MolmoBot-SPOC model trained on simulation data on the RB-Y1 platform, without any real robot data. See here for usage instructions.

Quickstart

import numpy as np
from huggingface_hub import snapshot_download
from molmobot_spoc.eval.config.rby1_eval_config import RBY1EvalBaseConfig
from molmobot_spoc.eval.config.spoc_policy_configs import SPOCRBY1RigidManipPolicyConfig
from molmobot_spoc.eval.spoc_policy import SPOCModelPolicy


ckpt_dir = snapshot_download("allenai/MolmoBot-SPOC-RBY1Rigid")
policy_config = SPOCRBY1RigidManipPolicyConfig(checkpoint_dir=ckpt_dir)
config = RBY1EvalBaseConfig(policy_config=policy_config, task_type="pick") # Try with "pick_and_place" too!

policy = SPOCModelPolicy(config, config.task_type)

obs = {
    "head_camera": np.zeros((576, 768, 3), dtype=np.uint8),
    "wrist_camera_r": np.zeros((576, 768, 3), dtype=np.uint8),
    "wrist_camera_l": np.zeros((576, 768, 3), dtype=np.uint8),
    "qpos": {
        "left_arm": np.zeros((7,), dtype=np.float32),
        "right_arm": np.zeros((7,), dtype=np.float32),
        "left_gripper": np.zeros((2,), dtype=np.float32),
        "right_gripper": np.zeros((2,), dtype=np.float32),
    },
    "goal": "pick up the apple",
}
# NOTE: get_action saves an internal buffer with the chunk and returns actions 1 by 1. To get the whole chunk, use model_output_to_action
action = policy.get_action(obs) 
print(action)

BibTeX

@misc{deshpande2026molmobot,
      title={MolmoB0T: Large-Scale Simulation Enables Zero-Shot Manipulation},
      author={Abhay Deshpande and Maya Guru and Rose Hendrix and Snehal Jauhri and Ainaz Eftekhar and Rohun Tripathi and Max Argus and Jordi Salvador and Haoquan Fang and Matthew Wallingford and Wilbert Pumacay and Yejin Kim and Quinn Pfeifer and Ying-Chun Lee and Piper Wolters and Omar Rayyan and Mingtong Zhang and Jiafei Duan and Karen Farley and Winson Han and Eli Vanderbilt and Dieter Fox and Ali Farhadi and Georgia Chalvatzaki and Dhruv Shah and Ranjay Krishna},
      year={2026},
      eprint={2603.16861},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2603.16861},
}