albagon/telearms_dr300_1look

TeleArms DR300 train, 1 look per demo

LeRobot v3 dataset (albagon/telearms_dr300_1look): bimanual YAM teleoperation demos from the TeleArms game, re-rendered in MuJoCo with visual domain randomization for sim-to-real transfer of a pi0.5 policy.

Episodes	12,324
Frames	6,400,217 (59.3 h at 30 Hz)
Tasks	47
Robot	bimanual_yam; 14-D state/action: left_joint1..6, left_gripper, right_joint1..6, right_gripper
Cameras	camera_front, camera_left (wrist), camera_right (wrist), 224x224, letterboxed from the real cameras' aspect ratios
Grippers	1.0 = open, 0.0 = closed
Action stats	relative (action - state for joints), pinned from helen9975/pi05_telearms_v4_fullft_lr7.07e-5@4a598baa0ecab581327eb77550820fdbc2b16341

• Training demos: up to 300 per task from the training split of checkpoint helen9975/pi05_telearms_v4_fullft_lr7.07e-5 (ckpt0); none come from ckpt0's holdout (checked against provenance/ckpt0_holdout_uuids.json).
• One DR look per demo (the __dr00 render). Same demos as albagon/telearms_dr300_2looks, which adds a second look.
• 11 episodes whose wrist view was near-black for the whole episode are excluded (provenance/exclude_dark_wrist_train.json).
• ckpt0's train/holdout split by demo UUID: provenance/ckpt0_split_by_uuid.csv. The DR shares in the table below are for the whole render; the dr_report summary covers all looks.

Domain randomization (observed shares)

Factor	Value	Configured	Observed
table source	procedural	0.20	0.200
table source	table_other	0.20	0.196
table source	table_real	0.40	0.404
table source	table_wood	0.20	0.200
lighting setup	lamp	0.10	0.102
lighting setup	original	0.15	0.145
lighting setup	panels	0.50	0.505
lighting setup	wide	0.15	0.149
lighting setup	window	0.10	0.099
floor replaced	False	0.10	0.100
floor replaced	True	0.90	0.900
skybox replaced	False	0.10	0.104
skybox replaced	True	0.90	0.896
procedural backdrop dark	False	0.40	0.395
procedural backdrop dark	True	0.60	0.605
JPEG applied (per camera)	False	0.50	0.501
JPEG applied (per camera)	True	0.50	0.499

Full config and per-factor ranges: provenance/dr_config_resolved.json; render report: provenance/report.html and provenance/summary.json.

Episodes per task

Task	Episodes
Adjacent Red Cube Placement	300
Align Blocks	299
Aligned Cube Row Placement	299
Aligned Letter Cube Row	144
Alphabet Cube Row Arrange	99
Alternate Green Yellow Row	256
Alternate Red Blue Row	299
Arrange Alphabetical Order	155
Arrange Shapes	300
Block Line Arrangement	207
Blue Base Red Pyramid	110
Blue Cube On Red Target	300
Blue On Red Stack To Green Target	300
Colored Cube Row Arrangement	299
Cube Center Placement	300
Cube Cluster Center	300
Cube Forward Right Placement	298
Cube Midpoint Placement	182
Cube Patch Placement	300
Cube Row Arrangement	259
Dog Cube Row Arrangement	190
Front Edge Cube Row	300
Front To Back Cube Line	300
Green Cube On Yellow Target	300
Hi Cube Placement	300
L Cube Arrangement	300
Letter Face Alignment	300
Microsoft Logo Cubes	300
Opposite Corner Cube Placement	200
Push Block to Target	300
Red Behind Blue Placement	300
Red Cube Center Placement	300
Red Cube In Front Of Blue Cube	300
Red Left Of Blue Placement	299
Red On Semantic Cube Stacking	299
Rotate Letter Block	300
Semantic Cube Blue B Placement	300
Semantic Cube Green Target Red Stack	300
Semantic Cube On Red Patch A Up	300
Semantic Cube Pyramid Build	240
Semantic Cubes 1x4 Line 1234	126
Sequential Push Block to Target	300
Spell Ok Cube Alignment	143
Three Cube Pyramid Stacking	283
Three Cube Wall Row	256
Top Letter Cube Stacking	183
Vertical Insert Two Blocks	299

Loading

from lerobot.datasets.lerobot_dataset import LeRobotDataset
ds = LeRobotDataset("albagon/telearms_dr300_1look")