Add artifact card
Browse files
README.md
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- robotics
|
| 4 |
+
- lerobot
|
| 5 |
+
- robotwin
|
| 6 |
+
- putcab
|
| 7 |
+
- preview
|
| 8 |
+
task_categories:
|
| 9 |
+
- robotics
|
| 10 |
+
---
|
| 11 |
+
# PutCab Concurrent: in-place waiting, Preview10
|
| 12 |
+
|
| 13 |
+
Ten qualified scene-matched demonstrations for user review. Three 320×240 H.264
|
| 14 |
+
camera views at 50/3 FPS; simulation runs at 250 Hz. Frames: 1641.
|
| 15 |
+
Left arm opens the drawer; right arm grasps and lifts the object at its original
|
| 16 |
+
location. Transfer and placement follow the required drawer readiness condition.
|
| 17 |
+
|
| 18 |
+
Sequential uses five A-then-B and five B-then-A demonstrations. Concurrent starts
|
| 19 |
+
both native expert programs at the same physics tick with independently advancing
|
| 20 |
+
control queues. Mixed selects five episodes from each parent on complementary
|
| 21 |
+
scene seeds. CTR uses the E098 V4 source trajectories, C2 transitions and selected
|
| 22 |
+
random deltas, with feedback waiting at the in-place safe-lift waypoint.
|
| 23 |
+
Ordinary expert episodes may use a front-clearance transit waypoint with
|
| 24 |
+
scene-specific approach height and orientation; each episode's transport strategy is recorded
|
| 25 |
+
in its lineage. Waiting remains at the original grasp-and-lift endpoint.
|
| 26 |
+
|
| 27 |
+
`observation.arm_active_mask` is float32[2], ordered left/right. At observation t,
|
| 28 |
+
it describes the outgoing action interval: 1 if the arm advances its task program
|
| 29 |
+
at any of the next 15 physics steps; 0 for pre-start, explicit holding, gate waiting,
|
| 30 |
+
completed programs and padding. Gripper commands and their programmed settling
|
| 31 |
+
dwell belong to the active program. The final repeated action has mask [0,0].
|
| 32 |
+
`observation.arm_motion_mask` separately reports whether any of the arm's seven
|
| 33 |
+
commanded coordinates changes by more than 1e-7 inside that same interval.
|
| 34 |
+
Measured physical joint motion is available in `meta/actual_joint_state`.
|
| 35 |
+
|
| 36 |
+
`meta/arm_progress` retains physics-step progress counters for independent mask
|
| 37 |
+
reconstruction. `meta/lineage.jsonl` records scene identities, command hashes,
|
| 38 |
+
mask receipts, scheduling events and parent revisions. These annotations are
|
| 39 |
+
part of the data artifact; model-side use is specified in separate training runs.
|
| 40 |
+
|
| 41 |
+
Each episode passed task success and full-scene penetration ≤10 mm in generation
|
| 42 |
+
and fresh exact-control video replay. All 30 videos were decoded and timestamps
|
| 43 |
+
checked; arm masks were reconstructed from the executed progress counters.
|
| 44 |
+
|
| 45 |
+
Experiment: E107-R001. Lifecycle: reported → audit pending.
|
| 46 |
+
Full 50-episode construction awaits user preview approval.
|