Model Card for molmoact2_mpc_200

MolmoAct2 fine-tuned on motion-planning demos for four MPC tabletop ManiSkill tasks (absolute end-effector pose / pd_ee_pose).

Fine-tuned from allenai/MolmoAct2-LIBERO using LeRobot.

Training Dataset

• Repository: jstm/mpc_lerobot_pd_ee_pose_200
• Control mode: absolute end-effector pose (pd_ee_pose)
• Cameras: camera_center, camera_left, camera_wrist (378×378)
• Total demonstrations: 767

Env	Demonstrations
PickCube-v2-wrist	192
PushCube-v2	192
LiftPegUpright-v2	191
PullCubeTool-v2	192

Model Details

• License: apache-2.0
• Base model: allenai/MolmoAct2-LIBERO
• Policy type: molmoact2
• Action mode: both (use --policy.inference_action_mode=continuous at eval)
• Setup type: single franka robotic arm
• Control mode: absolute end-effector pose
• Chunk size / n_action_steps: 10
• Training steps: 10000

Inputs & Outputs

Inputs

Feature	Type	Shape
observation.state	STATE	(9,)
observation.images.camera_center	VISUAL	(3, 378, 378)
observation.images.camera_left	VISUAL	(3, 378, 378)
observation.images.camera_wrist	VISUAL	(3, 378, 378)

Outputs

Feature	Type	Shape
action	ACTION	(7,)

Evaluation

Example eval on an MPC tabletop task:

lerobot-eval \
  --policy.path=jstm/molmoact2_mpc_200 \
  --policy.device=cuda \
  --policy.inference_action_mode=continuous \
  --trust_remote_code=true \
  --env.type=maniskill \
  --env.control_mode=pd_ee_pose \
  --env.task="PickCube-v2-wrist::lift the red cube" \
  --env.episode_length=100 \
  --env.observation_height=378 \
  --env.observation_width=378 \
  --env.goal_height=0.1 \
  --eval.n_episodes=100 \
  --eval.batch_size=20 \
  --output_dir=outputs/molmoact2_mpc_200__PickCube-v2-wrist