Predicting the upper bound of human path keeping performance using the two-point visual steering model: a comparison...

Area of Science:

• Cognitive science
• Human-computer interaction
• Robotics

Background:

• Previous models using ACT-R's three-step serial processing accurately predicted human path-keeping performance limits.
• The specific role of cognitive architecture versus processing steps in this predictive ability remained unclear.

Purpose of the Study:

• To investigate whether cognitive architecture or the number of processing steps is crucial for predicting human path-keeping performance upper bounds.
• To compare different computational models against human steering data.

Main Methods:

• Compared four implementations: no cognitive architecture, ACT-R, and two QN-MHP variants.
• Parameterized models to minimize average path-keeping error without fitting to human data.
• Validated model predictions against human steering performance.

Main Results:

• Implementations without cognitive architecture or separate two-point visual processing exceeded human performance, failing to predict realistic upper bounds.
• A two-step serial processing model within a cognitive architecture successfully predicted human steering performance upper bounds.

Conclusions:

• A cognitive architecture is necessary for realistic upper-bound predictions of human path-keeping.
• Two-step serial processing is both necessary and sufficient for maintaining predictive accuracy within a cognitive architecture.