Gallery · RoboDojo

stack bowls, layout 5

Task: “Stack the three bowls together.” DeepSeek-V4.1-Flash, code mode: 10 decisions, 57 motions, after step limit. The benchmark's check passed after its official step limit.

2× · 47 s of motion
  1. 1 of 10 · pick1.py
  2. 2 of 10 · put_down2.py
  3. 3 of 10 · regrasp.py
  4. 4 of 10 · place1.py
  5. 5 of 10 · release1.py
  6. 6 of 10 · pick2.py
  7. 7 of 10 · place2.py
  8. 8 of 10 · pick3.py
  9. 9 of 10 · place3.py
  10. 10 of 10 · verify.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. grasp_execute
  13. grasp_execute
  14. grasp_execute
  15. grasp_execute
  16. grasp_execute
  17. wait_ticks
  18. move_pose
  19. move_pose
  20. move_pose
  21. move_pose
  22. set_gripper
  23. wait_ticks
  24. move_pose
  25. wait_ticks
  26. return_home
  27. wait_ticks
  28. move_pose
  29. move_pose
  30. set_gripper
  31. wait_ticks
  32. move_pose
  33. wait_ticks
  34. move_pose
  35. move_pose
  36. move_pose
  37. move_pose
  38. move_pose
  39. set_gripper
  40. wait_ticks
  41. move_pose
  42. wait_ticks
  43. move_pose
  44. move_pose
  45. set_gripper
  46. wait_ticks
  47. move_pose
  48. wait_ticks
  49. move_pose
  50. move_pose
  51. move_pose
  52. move_pose
  53. set_gripper
  54. wait_ticks
  55. move_pose
  56. wait_ticks
  57. 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.