Stopping-Point Consistency
Raw data
We get a position and orientation in 3-space, at each barrier of each trial, taken at the time the user
# came to full stop in front of the dial,
# began spinning the dial,
# completed the dial.
Each position is measured relative to its respective dial-center, so everything is in the same coordinate space.
Per-barrier measures
(one for each subject x condition x trial x barrier)
- X and Y independently
- distance form dial-center (length of head-to-dial vector)
- incidence angle (in XY plane) of the head-to-dial vector
- angle between gaze direction and the head-to-dial vector. (probably just 1 angle, measured in the XY plane)
- angle between gaze direction and +y (1 angle in XY)
- angle between head-to-dial vector and +y (1 angle in XY)
Per-condition measures:
One for each subject * condition. Aggreggated over all trials * barriers. For each subject * condition, there is a cluster of 12 data points.
- cluster mean: X and Y values
- std along major axis (PC1)
- std along minor axis (PC2)
Velocity Profiles
- Forward velocity was most helpful, but we also could have used any of the other scalar per-sample signals: acceleration (forward, normal), curvature.
- Align the paths [at what mark?]
- Either stretch or crop to make them all the same length.
- Sample regularly throughout the range.
- Can compute mean and stdv for each point.
- Can do PCA to show stdv in terms of individual PCs.
Cluster plots for point-wise similarity
- This can be done for any scalar per-sample signal. Velocity was most helpful.
Cluster plots for per-path measures