Day 1, part 5: blinks yes, left and right no
In short
- With good contact: blinks 3 of 3, left and right 0 of 4.
- Two wrong assumptions in our own code, found in the saved recording and fixed.
- A Muse S forehead barely sees sideways eye movements, so we need sensors beside the eyes.
- Checkpoint 01 is tagged.
On this page
The setup
Forehead dampened, both forehead sensors at 100% contact, the Muse confirmed preset p21 (firmware 2.0.11). The eye check said: 0 of 3 blinks, 0 of 4 left/right. This time it saved its recording, so we could look.
Wrong assumption 1: when to measure “rest”
The check measured the resting wiggle in the first three seconds and got 46 µV. Later, in the second “eyes still” moment, it was 10–11 µV. People are still settling in when a test starts. Because every threshold scaled with the rest value, a blink had to exceed 184 µV.
Fix: use the quieter of the two “eyes still” moments.
Wrong assumption 2: which way a blink goes
We assumed a blink moves both forehead sensors in the same direction. On this Muse S, it didn’t:
| Blink at | AF7 | AF8 |
|---|---|---|
| 3 s | +303 µV | −465 µV |
| 5 s | +170 µV | −117 µV |
| 7 s | −72 µV | −46 µV |
Which way a sensor swings depends on where the headband measures against (its reference). The data was right; our rule wasn’t.
Fix: a blink counts when both sensors respond clearly, in any direction.
The real finding: left and right barely show
| Prompt | AF7 | AF8 |
|---|---|---|
| look left (11 s) | +51 µV | +18 µV |
| look right (12.5 s) | +7 µV | +2 µV |
| look left (14 s) | −8 µV | −8 µV |
| look right (15.5 s) | +1 µV | +47 µV |
Next to blinks of hundreds of microvolts, that’s practically nothing. With the corrected rules, the same recording scores 3 of 3 blinks, 0 of 4 left/right. The check still fails, and now it fails for the right reason.
Why? The forehead sensors of the Muse S sit close to the middle of the forehead (a validation study even names them Fp1 and Fp2). Sideways eye movements produce their biggest signal beside the eyes, which is exactly where sleep labs put their eye electrodes.
What this means
- A Muse S can probably see that your eyes move during REM sleep, and it clearly sees blinks.
- It probably cannot reliably see a left-right-left-right signal sent from inside a dream. That’s one person, one headband and one day, so we’ll test more, but it’s a measured result, not a guess.
- This is why we’re looking at a cheap, home-built eye sensor (EOG) with electrodes beside the eyes.
Lessons
- Save the data before you judge. The check failed three times; only the saved recording showed why.
- Measure, don’t assume. Both wrong rules came from what we expected a headband to do.
- A failing test can be the right answer. The point of the eye check is to tell the truth about the hardware, not to pass.
Checkpoint 01
With this, the first milestone is done and tagged as checkpoint-01: any EEG source, a Muse S over Bluetooth, Fake Moose, a live view with a contact score, and an eye check that tells the truth. Measured on our Muse S Gen 2: blinks yes, left and right no.
git clone https://github.com/emolionl/openlucid
cd openlucid
git checkout checkpoint-01
Next: a home-built eye sensor (EOG) beside the eyes, researched on the research page.