This paper presents a new machine learning method, called Light Dual-Schemata model. Dual-Schemata mod1 el is a framework for subjective symbol generation. Light Dual-Schemata model is a specialized version of a general Dual-Schemata model. In the context of machine-learning research, machine designers and/or task designers decide most problems for an agent to learn. In the future, however, they must find target concepts to learn thorough interactions with environments and/or other agents by themselves. Our Dual-Schemata model gives an autonomous agent an ability to notice differences among dynamic environments. This concept is inspired by Piaget's Schema model. Dual-Schemata model realizes apart of this cognitive development model as computational model. An experiment is shown as an actual example of the model. In this experiment an autonomous facial robot becomes able to chase each ball movements, to create symbols corresponding to environmental dynamics, and to recognize each movement, without any teaching signals. © 2004 IEEE.