callummole / sp_18-19 Goto Github PK

increase the frequency of calibrations.

reduce the size of the grid.

incorporate eye and steering synchronization alterations.

The original experiment (tag v1) had a factorial design, with four levels of steering angle bias (SAB; offsets to yaw-rate). Two of these were subtle and the participant did not leave the road (our 'No Failure' conditions). Only two of these consistently elicited take-overs (our 'Sudden' and 'Gradual' failure conditions).

The most promising avenue for modelling seems to be using time to (lane) crossing (TTC). This collapses across 'levels' of failure. So using only two values of rate of TTC change ('Sudden' and 'Gradual') risks overfitting a model.

An alternative option would to have a continuous measure of SAB, or at least a large number of levels.

The jury is currently out on whether we would want a large number of levels (e.g. 10), or a continuous measure which is sampled from in a quasi-random manner

Currently the yawrate offset is introduced suddenly after onset time X. This isn't very realistic and participants could be reacting to the sudden change in yawrate rather than any threshold of time-to-collision or some such.

It would be better if the yawrate bias would be gradually introduced via interpolation or some sort of smoothing method.

The steering wheel values were capped, meaning the steering signal at high SABs were contaminated. Increase sensitivity to counter this.