Relational reinforcement learning applied to shared attention

Abstract

This paper describes the design and implementation of a learning method in the context of robotic architecture for the social interactive simulation. This method is based on TG algorithm, named ETG, but use incremental process during the episode of learning. So, it does not use secondary memory to storage examples before insert in relational regression engine. This make easier the agent to choose the action with a greater degree of accuracy. The performance of ETG has been tested into a robotic architecture that control a head robotic. Then, a set of empirical evaluations has been conducted in the social interactive simulator for performing the task of shared attention. The experimental results show that the proposed algorithm is able to produce appropriate learning capability for shared attention.