Configure taxel patterns

Use GridPatternCfg for a flat pad, CustomPatternCfg for arbitrary NPZ poses, or Dex01PatternCfg for the DEX-01 exterior. Custom NPZ files contain poses of shape (N,3,4), with the third axis as the surface normal, and integer taxel2pixel pairs (row,column). dex01_pattern.npz preserves the corrected original layout; the exterior pattern supplied by Dex01PatternCfg contains the contact sampling poses. Their taxel order and pixel mapping are identical.

Read geometry and coordinate frames before changing a layout or mounting transform. Pattern configuration requires an initialized Isaac Sim application.

The same firmware rows color the exterior samples and frame locations. Grey pixels are unoccupied. These are layout coordinates, not force readings.

Configuration fields

Pattern	Required or supplied fields
GridPatternCfg	resolution and size in metres; ordering is xy by default or yx
CustomPatternCfg	NPZ filename and resolution in metres
Dex01PatternCfg	Supplied exterior NPZ and resolution (0.00072, 0.00088) m

The pattern generator returns positions (T, 3), axes (T, 3, 3), and pixel indices (T, 2). Use the matching mounting transform when selecting the supplied exterior pattern.

Validate a custom layout

Before loading a custom NPZ into a scene, verify the file arrays:

import numpy as np

with np.load("my_pattern.npz", allow_pickle=False) as layout:
    poses = layout["poses"]
    pixels = layout["taxel2pixel"]
assert poses.ndim == 3 and poses.shape[1:] == (3, 4)
assert pixels.shape == (len(poses), 2)
assert np.isfinite(poses).all()
assert np.issubdtype(pixels.dtype, np.integer) and (pixels >= 0).all()
assert len(np.unique(pixels, axis=0)) == len(pixels)
axes = poses[:, :, :3]
assert np.allclose(axes.transpose(0, 2, 1) @ axes, np.eye(3), atol=1e-6)
assert np.allclose(np.linalg.det(axes), 1, atol=1e-6)

Use local positions in metres and right-handed orthonormal axes. The third axis is the outward surface normal. Pixel indices start at zero and are ordered (row, column). Empty cells are automatically zero-filled; choose indices appropriate to the frame dimensions your consumer expects.