Design and performance of symbols self-organized within an autonomous agent interacting with varied environments

This work presents a novel machine learning model for autonomous agents. That is light dual-schemata model. Light dual-schemata model is a framework for subjective symbol formation. Robots equipped with light dual-schemata model can differentiate their concepts about environmental dynamics, which are called "perceptional schema". This differentiation comes out by a robot\´s subjective error estimates, rather than an objective error defined by a designer, which enables a robot\´s subjective differentiation process. An experiment is shown to prove its reasonableness. In the experiment, a facial robot forms appropriate schemas so as to chase a moving ball in a simulation world. This formation process deeply depends on the interaction context which is designed not by a designer who produced the robot, but by a caregiver who interacts with the robot.