How internal modeling arises when 'the world is not enough': an evolutionary robotics study

摘要

The aim of this study is showing that a simulated robot trained in a navigation task with a genetic algorithm can develop an internal model, and rely on it to fulfill the same task adaptively even in (partial) absence of external stimuli, or when the robot is temporarily 'blindfold'. We found that evolved internal models have dynamical and aticipatory aspects. In our experiments the key condition is unreliability in sensory stimulation.