Tianji RoboTwin Collection Task Status

Updated: 2026-07-10 UTC

Label Meaning

complete: reached or exceeded the current target of 100 valid HDF5 episodes.
slow_collectable: can produce valid HDF5 episodes, but throughput is low because many pre-motion successes fail during replay/save.
failed_or_unusable: monitored probes repeatedly hit 20 consecutive failures and 0 success, or produced no usable HDF5 in the probe run.
partial_unstable: some data exists, but later attempts were repeatedly unsuccessful; keep as partial unless specifically required.

Current valid HDF5 total under data/*: 1629 episodes.

Task	HDF5 Count	Label	Notes
shake_bottle	101	complete	Primary top task.
shake_bottle_horizontally	100	complete	Primary top task.
grab_roller	107	complete	Primary top task.
place_container_plate	100	complete	Primary top task.
click_alarmclock	104	complete	Primary top task.
adjust_bottle	105	complete	Primary top task.
open_laptop	108	complete	Primary top task.
place_empty_cup	106	complete	Extra task; intermittent failures, but reached target.
press_stapler	106	complete	Extra task; stable enough, usually saved several per round.
beat_block_hammer	106	complete	Extra task; reliable after it started collecting.
move_playingcard_away	102	complete	Newly found collectable task.
place_phone_stand	106	complete	Newly found collectable task.
place_object_stand	103	complete	Newly found collectable task.
place_object_scale	102	complete	Newly found collectable task.

Task	HDF5 Count	Label	Evidence / Notes
place_burger_fries	99	slow_collectable	Can collect, but replay/save yield is low. Many rounds saved only about 1-6 out of 10 pre-motion successes. Only 1 short of 100.
move_stapler_pad	56	slow_collectable	Newly found collectable task, but not yet enough data; continue if more data is needed.

Task	HDF5 Count	Label	Evidence / Notes
lift_pot	18	failed_or_unusable	Has old partial data, but repeated later attempts hit `20 consecutive failures and 0 success` for many rounds; should be marked failed/unstable for this setup.
stack_bowls_two	0	failed_or_unusable	Probe produced no HDF5; monitor terminated after 20 consecutive failures and 0 success.
move_can_pot	0	failed_or_unusable	Probe produced no HDF5; monitor terminated after 20 consecutive failures and 0 success.
pick_dual_bottles	0	failed_or_unusable	Probe produced no HDF5; monitor terminated after 20 consecutive failures and 0 success.
put_bottles_dustbin	0	failed_or_unusable	Probe produced no HDF5; monitor terminated after 20 consecutive failures and 0 success.
place_object_cabinet	0	failed_or_unusable	Probe produced no HDF5.
open_microwave	0	failed_or_unusable	Probe produced no HDF5; monitor terminated after 20 consecutive failures and 0 success.
stack_bowls_three	0	failed_or_unusable	Probe produced no HDF5; monitor terminated after 20 consecutive failures and 0 success.
rotate_qrcode	0	failed_or_unusable	Probe produced no HDF5; monitor terminated after 20 consecutive failures and 0 success.
dump_bin_bigbin	0	failed_or_unusable	Probe produced no HDF5; monitor terminated after 20 consecutive failures and 0 success.
handover_mic	0	failed_or_unusable	Probe produced no HDF5; monitor terminated after 20 consecutive failures and 0 success.
place_object_basket	0	failed_or_unusable	Probe produced no HDF5; monitor terminated after 20 consecutive failures and 0 success.

Probe new collectable tasks:

bash ./probe_new_collectable_tasks_epic.sh 20 20 3 5

Continue the newly found 5 tasks to 100:

bash ./collect_new5_to_100_epic.sh 100 20 20

For reporting:

Use failed_or_unusable for tasks that repeatedly failed during collection and should not be treated as merely slow.
Use slow_collectable for tasks that can produce data but need much more wall-clock time per valid episode.
Current useful completed set is strong enough for delivery; if more data is needed, prioritize move_stapler_pad and optionally the last missing place_burger_fries episode.