Add Melissa & Doug ABC/123 wooden block set

[siddhss5]

Goal

Add the full 50-block Melissa & Doug Wooden ABC/123 Blocks set to prl_assets.

Block specs

• 50 one-inch (25.4mm) cubes
• Wooden, hand-painted with letters, numbers, and pictures
• Colors: red, blue, green, yellow, natural wood backgrounds
• Mass: ~15g per block

MuJoCo properties

• Geom: box with half-extents [0.0127, 0.0127, 0.0127]
• Friction: 0.8 (wood-on-wood)
• Each block is a separate object type (block_A, block_B, ..., block_1, block_2, ...) or a single wooden_block type with texture variants

Texture pipeline

Step 1: Photograph the real blocks

• Photograph each face of each block (6 faces × 50 blocks = 300 photos)
• Use a consistent setup: white background, even lighting, overhead camera
• Name files: block_A_front.jpg, block_A_back.jpg, etc.

Step 2: Create UV-mapped textures

• Stitch the 6 face photos into a cube unwrap texture (cross layout)
• Script to automate: crop, align, compose into single PNG per block
• Output: block_A.png, block_B.png, etc.

Step 3: Create MuJoCo assets

• One XML per block (or one XML with material variants)
• meta.yaml with geometric properties (box, 0.0127m half-extents)
• Register textures as MuJoCo texture + material assets

Step 4: Verify

• Load all 50 in a scene, check grasping with Robotiq
• Verify stacking stability in physics mode
• Check that TSR grasp_box() works for the small size

Design questions

• One type or 50 types? A single wooden_block type with 50 instances (same physics, different textures) is simpler. But block_A, block_B etc. allows asking the robot "pick up the letter A" which is more interesting for demos.
• Grasp feasibility: 25.4mm is small for the Robotiq 2F-140 (min opening ~0mm, max 140mm). Should work but may need tuned grasp TSR tolerances.

Use cases

• Stacking demos (build a tower, spell a word)
• Sorting (by color, by letter vs number)
• LLM-driven tasks ("spell HELLO", "stack the vowels")