mliu48/telearms-yam-cube-v4-dr

TeleArms YAM cubes v4: camera-realistic, with colour and letter variants

The demonstrations of mliu48/telearms-yam-cube-v4 (48 bimanual YAM cube-manipulation tasks, MuJoCo), re-rendered the way the real cameras see them and extended with colour and letter variants. LeRobot v3.0, 30 Hz, about 330 GB.

	Episodes	Frames
Original teleoperated demos (48 tasks)	61,027	30,803,657
Synthetic variants (1,436 tasks)	38,469	19,810,359
Total	99,496	50,614,016

Cameras

Each camera renders at 640×360 (16:9) with its real field of view, then goes through pi0.5's resize_with_pad to 224×224: bilinear scaling, centred black bars, no crop.

Key	Real camera	Field of view (H × V)
observation.images.camera_front	RealSense D435IF colour	69° × 42°
observation.images.camera_left, camera_right	RealSense D405 (wrists)	87° × 56°

meta/info.json records camera_resolution: [640, 360]; feed real frames through the same letterbox. For pi0.5, the keys map to base_0_rgb, left_wrist_0_rgb and right_wrist_0_rgb.

Visual domain randomization

Every episode draws its own light position, direction, colour and strength; procedural table, floor and background textures in low-saturation colours, or the original ones (15 %); camera pose jitter (±2 cm and ±2° on top, ±0.5 cm and ±2° on the wrists); and field-of-view jitter of ±2°. Physics and actions are untouched.

Synthetic variants

A variant replays an original demo with its cubes, patches or letter faces swapped one-to-one for other colours (red, blue, green, yellow, purple, orange) or letters, and its prompt rewritten to match. The motion is unchanged, so the variant succeeds at its new task by construction. Symmetric letters (H I N O S X Z) swap only among themselves; word tasks swap to real words; alphabetical rows stay in order.

• task_name is <original task> (relabeled); each distinct prompt is one synthetic task.
• Colour-only synthetic tasks have at least 10 demos each; letter tasks at least 50.
• One or two combinations per original task never appear, for held-out evaluation.

Prompts

meta/task_prompt_variants.json (format 2) holds 50 paraphrases per original task and, for each synthetic task, its colour and letter mapping. Synthetic tasks reuse their original task's paraphrases with the mapping applied. The original instruction stays in every pool.

State and action

State and action are 14-dimensional: 6 arm joints plus 1 gripper per arm. Actions are absolute joint targets. meta/stats.json holds normalization statistics in the 50-step relative space the training recipe uses (arm deltas to the chunk anchor, absolute grippers); do not renormalize relative actions with absolute statistics.