Are gripper positions normalized to [0,1]? (actions.right_gripper.position / states.right_gripper.position)

#5
by shaneH2000 - opened

Hi, thanks for your great work! I wonder whether gripper values such as actions.right_gripper.position are intended to be normalized to the range [0, 1].

I found that in sim/basic_tasks/genie1/meta/episodes_stats.jsonl, episode 1708 reports a maximum states.right_gripper.position value of 4.56, which suggests it may not be normalized to [0, 1]. Could you clarify the intended gripper value convention and the recommended way to obtain a normalized gripper state?

Intern Robotics org

Dear shaneH2000,

Regarding the Genie1 robot, the minimum and maximum values for states.left_gripper.position and states.right_gripper.position are [0.0, 5.74], where 0.0 corresponds to the gripper being fully closed and 5.74 to fully open. You may apply min-max normalization to scale these values appropriately.

Best regards,
Yang Tian

2026.1.9

Nimolty changed discussion status to closed

Dear shaneH2000,

Regarding the Genie1 robot, the minimum and maximum values for states.left_gripper.position and states.right_gripper.position are [0.0, 5.74], where 0.0 corresponds to the gripper being fully closed and 5.74 to fully open. You may apply min-max normalization to scale these values appropriately.

Best regards,
Yang Tian

2026.1.9

Thank you for your reply. Could you provide the per-embodiment ranges (min/max) for every embodiment in the dataset? This will help me normalize the signals consistently across robots.

Intern Robotics org
edited 4 days ago

Dear shaneH2000,

For Genie1 robot, min, max is [0.0, 5.74], where 0.0 is closed, 5.74 is open.
For ARX Lift-2 robot, min, max is [0.0, 0.088], where 0.0 is closed, 0.088 is open.
For AgileX Split Aloha, min, max is [0.0, 0.10], where 0.0 is closed, 0.10 is open.
For Franka with panda hand, min, max is [0.0, 0.08], where 0.0 is closed, 0.08 is open.
For Franka with Robotiq-2f-85 gripper, min, max is [0.0, 1.0], where 0.0 is closed, and 1.0 is open.

Best Regards,
Yang Tian

2026.1.11

Sign up or log in to comment