A comparison between human micro-affordances and computational classification

This study aimed to assess how specific components of an action could be selected by a simple computational system. We performed an experiment to test associations between grasps (precision or power grip) and several objects. We then ran simulations using a naive bayes classifier to study to what extent it could reproduce participants’ choice. This classifier had two learning matrices containing objects’ size associated with a grip by means of our experiment. When receiving a new object’ size it computed the probability for each grip to be adapted. The highest probability was considered to represent which grip was associated with the object by the classifier. Results show that the classifier can reproduce participants’ choice depending on the size of its learning matrices, and can quickly select the right type of grip for a majority of trials, showing that micro-affordances (Ellis & Tucker, 2000) can be reproduced through naive bayesian classification.