Gallery · LIBERO-Pro

Open the middle layer of the drawer

libero_goal_swap, task 0 · seed 3. DeepSeek-V4.1-Flash, code mode: 7 decisions, 13 motions, solved.

2× · 37 s of motion
  1. 1 of 7 · view1.py
  2. 2 of 7 · view2.py
  3. 3 of 7 · step_pull.py
  4. 4 of 7 · step_grasp2.py
  5. 5 of 7 · step3.py
  6. 6 of 7 · step4.py
  7. 7 of 7 · step5.py
  1. move_to
  2. move_pose
  3. move_pose
  4. move_pose
  5. move_pose
  6. move_pose
  7. move_pose
  8. move_to
  9. move_pose
  10. set_gripper
  11. move_to
  12. move_to
  13. move_to
Decisions

The recording is the robot's own cameras at 2×; the time between motions, while the model thinks, is not filmed. The 3D replay is the camera's picture in depth after each decision, with the paths it planned, to turn and step through.