Gallery · RoboDojo

insert tubes, layout 9

Task: “Insert the three tubes into the rack one by one.” DeepSeek-V4.1-Flash, code mode: 18 decisions, 49 motions, after step limit. The benchmark's check passed after its official step limit.

2× · 47 s of motion
  1. 1 of 18 · pick.py
  2. 2 of 18 · place1.py
  3. 3 of 18 · place2.py
  4. 4 of 18 · place3.py
  5. 5 of 18 · place4.py
  6. 6 of 18 · place5.py
  7. 7 of 18 · run3.py
  8. 8 of 18 · run5.py
  9. 9 of 18 · run7.py
  10. 10 of 18 · run8.py
  11. 11 of 18 · run9.py
  12. 12 of 18 · run10.py
  13. 13 of 18 · run11.py
  14. 14 of 18 · run12.py
  15. 15 of 18 · final.py
  16. 16 of 18 · press.py
  17. 17 of 18 · recover1.py
  18. 18 of 18 · final2.py
  1. move_pose
  2. move_pose
  3. set_gripper
  4. move_pose
  5. move_pose
  6. move_pose
  7. move_pose
  8. move_pose
  9. move_pose
  10. move_pose
  11. set_gripper
  12. move_pose
  13. wait_ticks
  14. move_pose
  15. move_pose
  16. move_pose
  17. move_pose
  18. set_gripper
  19. move_pose
  20. move_pose
  21. move_pose
  22. move_pose
  23. move_pose
  24. set_gripper
  25. move_pose
  26. wait_ticks
  27. move_pose
  28. move_pose
  29. set_gripper
  30. move_pose
  31. move_pose
  32. move_pose
  33. move_pose
  34. set_gripper
  35. move_pose
  36. wait_ticks
  37. move_pose
  38. wait_ticks
  39. return_home
  40. wait_ticks
  41. set_gripper
  42. move_pose
  43. move_pose
  44. wait_ticks
  45. set_gripper
  46. move_pose
  47. move_pose
  48. wait_ticks
  49. return_home
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.