IMU Coordinate Frame (Proto0)
Proto0 carries a Bosch BMI270 6-axis IMU (3-axis accelerometer + 3-axis
gyroscope) on SPI. See fw/docs/proto0-board-pinout.md for its pin assignments
and fw/src/imu/imu.cpp for the driver integration (samples both accel and gyro
at 25 Hz off the BMI270 data-ready interrupt).
The frame is the raw chip frame
The BMI270 devicetree node
(fw/boards/others/rgb_sunglasses_proto0/rgb_sunglasses_proto0_nrf5340_cpuapp_common.dts)
has no mount-matrix and no rotation applied. The firmware therefore
consumes the sensor's axes exactly as the part is soldered on the board —
accel[0]/gyro[0] is the chip's physical +X, and so on. There is no software
remap anywhere in the pipeline, so these photos are the authoritative
description of what each axis means on the wearer's head. If a future board
revision rotates the part, either add a mount-matrix to the devicetree or
update this document (and the images).
Axes as worn
Coordinates are given for the glasses worn normally, head upright and level.
| Axis | Points toward… | Anatomical | Rotation about it (gyro) |
|---|---|---|---|
| +X (red) | the crown / straight up | superior | yaw — "no". Positive = turning left |
| +Y (green) | out the left temple | wearer's left | pitch — "yes". Positive = nose down |
| +Z (blue) | out the back of the head | posterior | roll — ear to shoulder. Positive = crown tips left |
This is a right-handed triad (X × Y = Z). Gyro output for each axis is the angular rate about the corresponding accelerometer axis, positive per the right-hand rule.
Annotated views
In every photo the same triad is drawn — red = X, green = Y, blue = Z. The straight arrow(s) lie in the image plane; the curved arrow is the axis pointing roughly along the camera direction (toward or away from the viewer), drawn as an arc because it would otherwise be foreshortened to a dot.
Front — facing the wearer:
Back — behind the wearer, showing the rear electronics module:
Left side of the wearer:
Right side of the wearer:
Top-down — looking down at the crown, eyes at the bottom of the frame pointing at the floor:
Top-up — the same top view with the head tipped back the other way:
Bench verification
Both the accelerometer frame and the gyro polarity were verified on Proto0
hardware (2026-08-24, fw v3.4.0-stable) with the on-device capture path —
capture start <seconds>, whose .csv sidecar logs raw ax,ay,az,gx,gy,gz at
25 Hz. The axis directions above are measured, not inferred.
Accelerometer. The glasses were held stationary in six orientations; the axis pointing up reads +1 g. Every reading matched the table (m/s²):
| Orientation (as worn) | ax | ay | az |
|---|---|---|---|
| Upright, level | +9.95 | +0.83 | −0.01 |
| Upside down | −9.61 | −0.49 | −0.72 |
| Right temple down | −0.10 | +9.78 | −0.14 |
| Left temple down | −0.18 | −9.84 | −0.52 |
| Face down | +0.01 | +0.21 | +9.87 |
| Face up | +0.29 | −0.19 | −9.84 |
Off-axis terms are hand-held pose error (≤ 6°); every vector magnitude fell in 9.65–9.98 m/s², i.e. 1.00 g.
Gyroscope. Measured directly by sweeping briskly in the named direction and
returning slowly: gx peaked +6.6…+8.2 rad/s on left yaws, gy +10.6…+13.9 on
nose-down pitches, gz +5.4…+6.6 on left rolls — each with the slow return
reading the opposite sign.
Those signs were then cross-checked against the accelerometer's own gravity reading — a test that does not depend on the motions having been performed as intended — agreeing with median cosine similarity 0.92 (X), 0.91 (Y), 0.89 (Z). The gyro triad is therefore right-handed with respect to the accelerometer axes above, which pins the polarity by the right-hand rule.
Method, and the two ways it silently produces a false negative: fw/CLAUDE.md,
"Validating the IMU coordinate frame on hardware".