Learning to Avoid Collisions: A Functional State Space Approach

Two experiments examined performance in collision avoidance situations. In both experiments participants were asked to initiate a discrete maneuver to avoid a collision at the last possible moment. The affordances of the situations were varied as a function of vehicle dynamics and the functional consequences associated with responding too late or too early. The results were examined in the context of a 2-dimensional functional state space with dimensions associated with optical angle and optical expansion rate. The patterns of performance showed that the actions were consistent with decision rules that could be specified in terms of linear functions of the two optical variables. In most cases, performance at early stages of learning suggested that people were using an Expansion Rate Criterion. With practice, people would tune to a decision rule that was appropriate for the specific vehicle dynamics tested. The results are discussed in relation to the role of three factors in shaping ultimate performance: (a) tasks constraints (i.e., affordances), (b) information constraints (i.e., optical structure), and (c) experience (i.e., learning).