Adaptive acquisition with OpenTac-Xampling¶

OpenTac-Xampling reduces the number of taxel reads required per frame by concentrating measurements around contact. A coarse-to-fine search identifies taxels above a threshold, then expands into their neighbours. This trades complete spatial sampling in each frame for a higher update rate when contact occupies only part of the sensor.

Two 30 cm tennis-ball drops at 128× slow motion: raster scanning misses the impact; Xampling at XAMP_M = 32 captures it in this example. The displayed 996 updates/s is the measured rate of that recording.

Enable adaptive acquisition¶

Measure the unloaded baseline before enabling Xampling. The following commands select either a search budget with contact expansion or a fixed total sampling budget:

uv run opentac view --xampling 64             # 64 search reads, then grow around every contact
uv run opentac view --xampling 128 --fixed    # exactly 128 reads per frame, for a steady frame rate

from opentac.device import Sensor

s = Sensor()
s.zero()                       # nothing touching: measures each taxel's zero and noise
s.xampling(budget=128)         # XAMP_M = 128, TAC_MODE = xamp_grow (grow=False for xamp_fixed)
frame = s.read()
frame.mask                     # True for the taxels read this frame
frame.taxels                   # counts above the zero; NaN where not read
s.raster()                     # back to reading every taxel

In the viewer, taxels not read this frame are drawn blue, and b zeroes the board again.

Search budget versus total budget¶

Mode	`TAC_MODE`	Reads per frame
grow (default)	`xamp_grow` (2)	`XAMP_M` search reads, plus every taxel growth reaches: expansion continues over the detected contact beyond the search budget
fixed	`xamp_fixed` (1)	exactly `XAMP_M`, growth included: a steady frame rate, but contact coverage is incomplete if it requires more reads than the budget

Both modes reserve at least XAMP_SEARCH_MIN reads for searching before expanding detected contacts. This prevents previously detected contact from consuming the entire frame budget. Contact taxels found in one frame are prioritised in the next.

Settings¶

Parameter	Default	Meaning
`XAMP_M`	128	the budget: reads per frame (fixed) or search reads (grow)
`XAMP_THRESH`	30	counts above a taxel's own noise that count as contact
`XAMP_SEARCH_MIN`	30	search reads before any growth
`XAMP_ZERO_FRAMES`	64	raster frames the zero measures noise over
`XAMP_BUMP_FILL`	0	1: growth reads each contact taxel's whole shear bump
`XAMP_CONFIRM_US`	50	a reading that would trigger is read again after this extra settle

A taxel triggers when its baseline-subtracted reading exceeds its measured noise plus XAMP_THRESH. Restarting the board, changing the sensor, or changing ROW_SETTLE_US, MUX_SETTLE_US or gain invalidates this baseline. Xampling pauses until the baseline is measured again.

Measured frame rates¶

Fixed-mode measurements on a 32 × 32 sensor (1024 taxels), MUX_SETTLE_US 8, IMU off, including USB transfer. Two rounds each, averaged (2026-09-14).

`XAMP_M`	16	32	64	128	256	512	1024	raster
frames/s	1286	954	627	376	209	113	58	67
Relative to raster	19x	14x	9.3x	5.6x	3.1x	1.7x	0.9x	1x

Grow mode has no single frame rate: with nothing touching it reads XAMP_M taxels like fixed mode, and each taxel that growth reads adds roughly 13 to 16 µs. With the IMU on, its reads take about 40% of the frame time.

In grow mode at XAMP_M 128, a new 2 × 2 press was reached by the search after 1.1 frames on average and 7 frames at worst (32 × 32 sensor, hands-off measurement of the search pattern).