Vision-based Learning and Development for Emergence of Robot Behaviors

Abstract

: This paper focuses on two issues on learning and development; a problem of state-action space construction, and a scaling-up problem. The former is mainly related to sensory-motor mapping and its abstraction, and we show two our methods for the state and action space construction for reinforcement learning. For the latter issue, we attempt to dene the environmental complexity based on the relationships between observations and self motions. Based on this view, we introduce a method which can cope with the complexity of multi-agent environment by a combination of a state vector estimation process and a reinforcement learning process based on the estimated vectors. As example tasks in our work, we adopt the domain of soccer robots, RoboCup [1]. Computer simulations and real robot experiments are given. 1. Introduction The ultimate goal of our research is to design the fundamental internal structure inside physical entities having their bodies (robots) which can emerge complex behavior...