Eval sets: MP tasks record only head cameras (front_stereo videos and their info/stats entries removed), BendPickAndPlaceOnSofaMP videos re-encoded to h264, drop BendPickAndPlaceOnSofaMP dr-level-hard-old and BendPickAndPlaceTeleop videos.pre-yaw0
The policy instruction comes from environment_config.dr_state_dict.language, which held strings that differ from the training data:
- OpenFaucetTeleop: 'move forward to the faucet and open it' -> 'turn the faucet'
- XMovePickTeleop: 'bend to pick up the {}' -> 'move forward to pick up the {}'
Also aligned dr_cfgs.language.instructions, episode 'tasks' and meta/tasks.jsonl. No other fields changed.
