Gallery · RoboTwin

place dual shoes, seed 100001

Task: “Use both arms to move two the brown footwear with smooth texture to the orange box for keeping shoes, tips pointing left.” DeepSeek-V4.1-Flash, code mode: 14 decisions, 37 motions, solved.

2× · 60 s of motion
  1. 1 of 14 · step1_grasp_left.py
  2. 2 of 14 · step2_rotate_left.py
  3. 3 of 14 · step3_align_left.py
  4. 4 of 14 · step4_align_left2.py
  5. 5 of 14 · step5_place_left.py
  6. 6 of 14 · step6_checkleft_planright.py
  7. 7 of 14 · step7_right_grasp_align.py
  8. 8 of 14 · step8_right_rotate.py
  9. 9 of 14 · step9_place_right.py
  10. 10 of 14 · step10_open_retract.py
  11. 11 of 14 · step11_settle.py
  12. 12 of 14 · step12_home.py
  13. 13 of 14 · step13_settle.py
  14. 14 of 14 · step14_swapA.py
  1. grasp_execute
  2. grasp_execute
  3. grasp_execute
  4. grasp_execute
  5. grasp_execute
  6. move_to
  7. move_to
  8. rotate_wrist
  9. rotate_wrist
  10. move_pose
  11. set_gripper
  12. wait_ticks
  13. move_to
  14. move_to
  15. grasp_execute
  16. grasp_execute
  17. grasp_execute
  18. grasp_execute
  19. grasp_execute
  20. move_to
  21. rotate_wrist
  22. rotate_wrist
  23. move_pose
  24. set_gripper
  25. wait_ticks
  26. move_to
  27. set_gripper
  28. wait_ticks
  29. move_to
  30. recover_pose
  31. wait_ticks
  32. wait_ticks
  33. wait_ticks
  34. wait_ticks
  35. grasp_execute
  36. grasp_execute
  37. grasp_execute
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.