Extracting logical perceptual space for robot learning using factor analysis

Factor analysis has been employed for data analysis in behavioral sciences for decades. In this paper, we propose to employ it in robot behavior studies so that important underlying factors that affect the decision-making in robot behavior actions can be extracted. Causal relationships among physical (observed) and logical (unobserved) perceptual dimensions are constructed. Factor analysis provides a simple mean for us to understand what the sensors data, that construct the robot behavioral perceptual space S, are measuring (logical perceptual space extraction). Learning can thus be conducted based on the logical dimensions of the perceptual space, which usually has much lower dimensionality than the original physical perceptual space, of robot behaviors. Analysis of simulated obstacle avoidance behavior is presented