John Bear: Indistinguishable Gestures
Another participant, John Bear, exhibited a very different approach by repeatedly rerecorded samples to make them as similar as possible. They created jump and then added kick. They quickly found that these two gestures interfered with one another. In John Bear’s Gesture Log, notice that they removed and added kick six times. The gesture was interfering with jump due to similarities between the gestures’ accelerometer plots.
Different Gestures that Look Similar
When Brian Bear described their process for recording three different sleep samples, they explained that it was based on how they sleep, which looks different throughout the night.
What a Child Sees
When a child acts out a gesture with a stuffed animal, they see very different movements.
What a Computer Sees
The micro:bit detects acceleration along the x-, y-, and z-axis, and the resulting shape is compared against what the child’s future movements with the stuffed animal. In the case of sleep and jump, the computer sees both of these gestures very similarly.
When this occurred, their initial response was to add more samples for kick and jump. This revealed that they were aware that the number of samples affects the model. After recording a new jump sample, they inspected the existing samples, counted the three peaks, and then proceeded to record another sample, counting aloud as they moved the bear up and down three times. This again shows their care to make the samples similar while interpreting the sensor data to match peak count with the number of jumps.
John Bear’s Gesture Log
While making their project, John Bear discovered a problem: two of their gestures (kick and jump) were not triggering properly when tested.
John Bear iterated on their design, adding and removing kick recordings, in an attempt to fix this. They explained that the computer saw kick and jump very similarly, and would become confused.
Next: Findings