Testing the Tau Hypothesis against Alternative Information Sources used for Collision Avoidance

Research on the visual information used in collision avoidance has inappropriately attributed adequate task performance to the use of the visual variable tau (τ). We constructed a collision avoidance experiment where the τ hypothesis was tested against the use of other potential information sources, including expansion rate. Subjects viewed a head-on flight towards one or two barriers. During the first half of trials, a single barrier was present and subjects were instructed to approach it as closely as possible before ascending over it. During the second half of the trials, a second barrier was placed above the first and subjects flew through the resulting gap. The ascent dynamics were manipulated between subjects, such that a time-to-contact margin demarcated the collision boundary with one group, but a constant distance-to-contact margin with another group. Subjects in the time-relevant group responded in a manner more consistent with the use of expansion rate information, but neither a single critical expansion rate or τ strategy adequately explains performance in the two-barrier condition. Subjects in the distance-relevant group responded in a manner consistent with the use of relative distance information.

On Probability Models and Avoidance Learning

In the early 1970s Gibbon fitted empirical data from a number of avoidance experiments to a mathematical model based on the formalism of semi-Markov chains. Cotton and Wood in 1982 demonstrated the possibility that a single stochastic process underlies the timing behavior of rats in an unsignalled shuttlebox-avoidance learning experiment, providing quantitative support for Gibbon's hypothesis. This paper presents a further analysis in terms of semi-Markov chains of experimental data from an avoidance experiment with a variable, limited effective avoidance interval. The agreement between experiment and theory further supports that unsignalled avoidance experiments are convincingly modelled by semi-Markov chains.