Fxchennnn/aico_1_2026_05_21

RAIDS pi0.5 LeRobot dataset

This dataset is converted from already-collected RAIDS raw demos.

Learned and preprogrammed stages

• Learned by pi0.5: [2, 4, 5, 6, 7, 8]
• Excluded because they are preprogrammed tool routines: [1, 3]

Camera mapping

• front -> base_0_rgb
• left -> left_wrist_0_rgb
• right -> right_wrist_0_rgb

State/action schema

State and action are 16-dimensional float32 vectors:

• 0: left_arm_joint_0
• 1: left_arm_joint_1
• 2: left_arm_joint_2
• 3: left_arm_joint_3
• 4: left_arm_joint_4
• 5: left_arm_joint_5
• 6: left_arm_joint_6
• 7: right_arm_joint_0
• 8: right_arm_joint_1
• 9: right_arm_joint_2
• 10: right_arm_joint_3
• 11: right_arm_joint_4
• 12: right_arm_joint_5
• 13: right_arm_joint_6
• 14: left_general_grip_ratio
• 15: right_general_grip_ratio

Actions are absolute next-frame targets. The final two dimensions are simplified XHand grip ratios.