Gallery · LIBERO-Pro

Put the wine bottle on the rack

libero_goal_swap, task 9 · seed 2. DeepSeek-V4.1-Flash, code mode: 13 decisions, 42 motions, solved.

2× · 119 s of motion
  1. 1 of 13 · hover.py
  2. 2 of 13 · grasp_step.py
  3. 3 of 13 · grasp2.py
  4. 4 of 13 · place2.py
  5. 5 of 13 · regrasp1.py
  6. 6 of 13 · release1.py
  7. 7 of 13 · top_place.py
  8. 8 of 13 · bodygrasp.py
  9. 9 of 13 · grasp_lying.py
  10. 10 of 13 · lift_body.py
  11. 11 of 13 · grasp_plan2.py
  12. 12 of 13 · grasp_plan3.py
  13. 13 of 13 · place_slats.py
  1. move_pose
  2. move_pose
  3. set_gripper
  4. move_pose
  5. set_gripper
  6. move_pose
  7. move_pose
  8. set_gripper
  9. move_pose
  10. move_pose
  11. set_gripper
  12. move_pose
  13. wait_ticks
  14. move_pose
  15. move_pose
  16. set_gripper
  17. move_pose
  18. set_gripper
  19. move_pose
  20. wait_ticks
  21. set_gripper
  22. move_pose
  23. move_pose
  24. set_gripper
  25. move_pose
  26. set_gripper
  27. move_pose
  28. wait_ticks
  29. move_pose
  30. move_pose
  31. move_pose
  32. move_pose
  33. set_gripper
  34. move_pose
  35. wait_ticks
  36. set_gripper
  37. move_pose
  38. move_pose
  39. set_gripper
  40. move_pose
  41. wait_ticks
  42. move_pose
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.