Evaluating Model that Predicts When People will Speak to a Humanoid Robot and Handling Variations of Individuals and Instructions

We have tackled the problem of predicting when a user is likely to begin speaking to a humanoid robot. The generality of the prediction model should be examined so that it can be applied to various users. We present two empirical evaluations demonstrating that 1) our proposed model does not depend on the specific participants whose data were used in our previous data collection, and 2) the model can handle variations of individuals and instructions. We collect a data set to which 25 human participants in general public gave labels indicating whether or not they would be likely to begin speaking to the robot. We then train a new model with the collected data and determine its performance by cross validation and open tests. We also investigate the relation of how much individual participants feel likely to speak with a model parameter and the instruction given before data collections. Results show that our model can handle the variations.