Characteristics of learning and sensory dimension in two-mobile robots system

We investigate quantitatively the characteristics of relationships between learning and the sensory dimension. In the task of a two-mobile robotic system composed of a driving body and 1-DOF prismatic arm, carrying a long bar horizontally along a sinusoidal wavy road while keeping the bar-support length constant, a sensory-motor coordination is acquired by the reinforcement learning. We find the following results: when the two-mobile robotic system executes the task, a cooperative behavior performed by the forward and backward locomotions is needed, and we find 8 sensory dimensions that are necessary and sufficient in perception of the state for the robotic system. The sensory contribution rate of each sensory dimension for the perception of the robotic state is defined. Also, the two-mobile robotics system which is only equipped with sensors of large sensory contribution rate, can acquire the robotic control for the cooperative behavior of forward and backward locomotions by the online learning