galafinger commited on
Commit
eeaba2e
·
verified ·
1 Parent(s): 893bb19

Card: teaser, layout, limitations, privacy, citation, per-task duration, tags

Browse files
Files changed (1) hide show
  1. README.md +64 -9
README.md CHANGED
@@ -8,10 +8,17 @@ tags:
8
  - bimanual
9
  - human-demonstration
10
  - data-glove
 
 
 
 
 
 
 
11
  configs:
12
  - config_name: default
13
  data_files: data/*/*.parquet
14
- pretty_name: galafinger-manipulation-sample
15
  size_categories: 10K<n<100K
16
  ---
17
 
@@ -19,6 +26,10 @@ size_categories: 10K<n<100K
19
 
20
  Bimanual human demonstrations recorded with two **TouchTronix V3 thin-film tactile gloves** (left + right): fingertip pressure maps (5 × 4×4 per hand), finger bend, wrist orientation and automatic grasp-phase labels, both hands on one time axis. LeRobot v2.1 format; `meta/modality.json` lets NVIDIA Isaac GR00T read it directly.
21
 
 
 
 
 
22
  ## At a glance
23
 
24
  | | |
@@ -34,14 +45,14 @@ Bimanual human demonstrations recorded with two **TouchTronix V3 thin-film tacti
34
 
35
  ## Tasks
36
 
37
- | # | Instruction | Episodes |
38
- |---|---|---|
39
- | 0 | Pick the stones of two chosen colors from the mixed box and sort them onto two plates by color | 5 |
40
- | 1 | Squeeze the soft toy gently, then medium, then hard, and release | 10 |
41
- | 2 | Pick up the vials from the table one by one and place them upright into the slots of the tray | 5 |
42
- | 3 | Hold the cucumber on the cutting board and slice it into rounds with the knife | 5 |
43
- | 4 | Take the glasses out of the case, wipe both lenses with the cloth, and put the glasses back into the case | 6 |
44
- | 5 | Spread the T-shirt flat on the table and fold it into a neat rectangle | 6 |
45
 
46
  ## Hardware
47
 
@@ -76,6 +87,38 @@ Bimanual human demonstrations recorded with two **TouchTronix V3 thin-film tacti
76
 
77
  Egocentric stereo head camera (Orbbec EGO, left+right) + two tactile gloves, one operator, one room. Camera-glove sync is software-based: measured 0.1-0.2 s, not millisecond-level. Fingertip sensors report normal force only (no shear, no palm). Subtask labels were made from video plus force signals and checked against force edges; boundary accuracy about 0.25-0.5 s. The operator's clothing shows a brand wordmark in some frames. Small sample: 37 episodes, about 14 minutes.
78
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
79
  ## Loading data
80
 
81
  With LeRobot:
@@ -135,6 +178,18 @@ release 717- 747 peak 1.8 N
135
 
136
  Not in this release: hand or wrist position in space (the gloves give finger flex and orientation only), shear force, palm pressure, camera IMU.
137
 
 
 
 
 
 
 
 
 
 
 
 
 
138
  ## License and contact
139
 
140
  cc-by-4.0. For commercial licensing or custom collection (tasks, scenes, hours) contact Galafinger Labs: hello@galafinger.com · https://galafinger.com
 
8
  - bimanual
9
  - human-demonstration
10
  - data-glove
11
+ - egocentric-video
12
+ - stereo
13
+ - manipulation
14
+ - tactile-sensing
15
+ - force
16
+ - imitation-learning
17
+ - robot-learning
18
  configs:
19
  - config_name: default
20
  data_files: data/*/*.parquet
21
+ pretty_name: Galafinger Manipulation Sample
22
  size_categories: 10K<n<100K
23
  ---
24
 
 
26
 
27
  Bimanual human demonstrations recorded with two **TouchTronix V3 thin-film tactile gloves** (left + right): fingertip pressure maps (5 × 4×4 per hand), finger bend, wrist orientation and automatic grasp-phase labels, both hands on one time axis. LeRobot v2.1 format; `meta/modality.json` lets NVIDIA Isaac GR00T read it directly.
28
 
29
+ <video controls muted loop playsinline width="100%" src="https://huggingface.co/datasets/galafinger/galafinger-manipulation-sample/resolve/main/videos/chunk-000/observation.images.head/episode_000005.mp4"></video>
30
+
31
+ *Episode 5, head camera (left eye): squeeze the soft toy gently, then medium, then hard. The three force levels look alike on video and separate clearly in the fingertip force signal.*
32
+
33
  ## At a glance
34
 
35
  | | |
 
45
 
46
  ## Tasks
47
 
48
+ | # | Instruction | Episodes | Duration |
49
+ |---|---|---|---|
50
+ | 0 | Pick the stones of two chosen colors from the mixed box and sort them onto two plates by color | 5 | 3 min 07 s |
51
+ | 1 | Squeeze the soft toy gently, then medium, then hard, and release | 10 | 3 min 34 s |
52
+ | 2 | Pick up the vials from the table one by one and place them upright into the slots of the tray | 5 | 1 min 23 s |
53
+ | 3 | Hold the cucumber on the cutting board and slice it into rounds with the knife | 5 | 1 min 46 s |
54
+ | 4 | Take the glasses out of the case, wipe both lenses with the cloth, and put the glasses back into the case | 6 | 2 min 13 s |
55
+ | 5 | Spread the T-shirt flat on the table and fold it into a neat rectangle | 6 | 1 min 39 s |
56
 
57
  ## Hardware
58
 
 
87
 
88
  Egocentric stereo head camera (Orbbec EGO, left+right) + two tactile gloves, one operator, one room. Camera-glove sync is software-based: measured 0.1-0.2 s, not millisecond-level. Fingertip sensors report normal force only (no shear, no palm). Subtask labels were made from video plus force signals and checked against force edges; boundary accuracy about 0.25-0.5 s. The operator's clothing shows a brand wordmark in some frames. Small sample: 37 episodes, about 14 minutes.
89
 
90
+ ## Repository layout
91
+
92
+ ```
93
+ data/chunk-000/episode_000000.parquet … episode_000036.parquet glove streams + labels, one row per video frame (30 Hz)
94
+ videos/chunk-000/observation.images.head/episode_0000NN.mp4 head camera, left eye, 1600x900, 30 fps
95
+ videos/chunk-000/observation.images.head_r/episode_0000NN.mp4 head camera, right eye
96
+ meta/info.json feature names, shapes, fps
97
+ meta/tasks.jsonl task instructions
98
+ meta/episodes.jsonl episode -> task, length
99
+ meta/subtasks.jsonl sub-step label vocabulary
100
+ meta/annotations.jsonl sub-step time ranges per episode
101
+ meta/modality.json modality map (GR00T)
102
+ meta/stats.json, meta/episodes_stats.jsonl normalization statistics
103
+ ```
104
+
105
+ Episodes by task: 0–4 stones, 5–14 squeeze, 15–19 vials, 20–24 cucumber, 25–30 glasses, 31–36 T-shirt.
106
+
107
+ ## Limitations
108
+
109
+ - **Small.** 37 episodes, 13.7 minutes, one operator, one room, one table. Useful for inspecting the format and signal quality, not for training a policy on its own.
110
+ - **Sync is software-based.** Camera and gloves share a host clock; measured alignment is 0.1–0.2 s. Fine for grasp/hold/release timing, not for contact-onset events at millisecond scale.
111
+ - **Normal force only.** No shear force, no palm sensing, no slip signal.
112
+ - **No hand position.** The gloves give finger flex and orientation. Joint angles are estimated from one flex sensor per finger; there is no wrist or fingertip position in space, no depth, no camera IMU, no camera calibration in this release.
113
+ - **Sensor rest level.** Some fingertip sensors read 1–2 N with nothing in the hand; gentle contacts sit close to that level.
114
+ - **Labels.** Sub-step labels were drafted with AI assistance from video plus force and are accurate to about 0.25–0.5 s; they have not all been reviewed by a second person. Grasp-phase columns are automatic.
115
+ - **Repaired frames.** 12 corrupted video frames (one each in 12 videos) were replaced by the preceding frame; frame counts are unchanged.
116
+ - **Splits.** No predefined split. Hold out whole episodes; with a single operator and scene, results will not show generalization to new people or places.
117
+
118
+ ## Privacy
119
+
120
+ The only person recorded is the operator, a member of Galafinger Labs, who consented to publication. The camera looks down at the table and hands. Videos were machine-scanned frame by frame for corruption and reviewed visually at one frame per second; no faces, screens or documents were found. A brand wordmark on the operator's clothing is visible in some frames. If you spot anything that should not be public, write to hello@galafinger.com.
121
+
122
  ## Loading data
123
 
124
  With LeRobot:
 
178
 
179
  Not in this release: hand or wrist position in space (the gloves give finger flex and orientation only), shear force, palm pressure, camera IMU.
180
 
181
+ ## Citation
182
+
183
+ ```bibtex
184
+ @misc{galafinger2026manipulationsample,
185
+ title = {Galafinger Manipulation Sample: bimanual tactile-glove demonstrations with egocentric stereo video},
186
+ author = {{Galafinger Labs}},
187
+ year = {2026},
188
+ howpublished = {Hugging Face},
189
+ url = {https://huggingface.co/datasets/galafinger/galafinger-manipulation-sample}
190
+ }
191
+ ```
192
+
193
  ## License and contact
194
 
195
  cc-by-4.0. For commercial licensing or custom collection (tasks, scenes, hours) contact Galafinger Labs: hello@galafinger.com · https://galafinger.com