aw_mix_ee_hc

One cell of the 3-axis AI-Worker comparison grid (aw_<source>_<representation>_<cameras>).

• Source: co-training merge: robot exo data (episodes 0-59) + UMI VR data (episodes 60-105), 106 eps / 37,153 frames
• State/Action representation: 16-dim EE pose (left_ee/right_ee xyz + quaternion(x,y,z,w) + gripper) in the robot base frame. Robot-sourced poses were obtained by FK on rcv_worker.urdf with lift_joint=-0.15, TCP frames left_tcp/right_tcp. Gripper dims are identically 0 in both sources (never actuated).
• Cameras: head, wrist_left_cur, wrist_right_cur (240x320 @ 15 fps)

All variants share identical camera keys and fps, so any aw_*_ee_hc datasets are schema-compatible for co-training. Head-pose dims and vr_seq from the original UMI recording were dropped; head/lift joints from the original robot recording were dropped (lift is fixed at -0.15 in all kinematic conversions).

Built with LeRobot (dataset v3.0).