How To Bring Up A One-Eyed Mobile Robo-Infant
Takayuki Nakamura, Minoru Asada · 1995
An animat, a simulated animal or a real robot, has to learn a sequence of actions which reflect the dynam-ic interplay between the animat and its environment, mediated through the animat’s sensors and actuators. This paper presents a method to bring up an animat, a one-eyed mobile robo-infant in such a way that it learns the sensorimotor apparatus first (in its baby-hood), and then how to detect and avoid obstacles and how to pursuit a target (in its childhood), and finally how to combine these behaviors (in its youth). Key issues dealt with in this paper are twofold: • The robo-infant has visual tracking routines ca-pable of realtime acquisition of optical flow (vi-sual motion patterns) by which it can obtain the sensorimotor apparatus, and target pursuit and obstacle avoidance behaviors based on reinforce-ment learning. • The system’s (parents’) direct training (program-ming or teaching how to do) is avoided. Instead, the system changes the environments so as to make the robo-infant learn several behaviors to accomplish some tasks step by step. Computer simulation and real experiments are given to show the validity of the method. 1