A human-like learning approach to developmental robotic reaching

This paper presents a human-like approach for robot to obtain reaching ability autonomously in three-dimensional environment. The essential elements of the approach are inspired by current findings in neural science research and developmental psychology. By imitating the mechanism of the infant realizing the body induction, the robot system realizes the automatic separation of the mechanical arm and the external environment. We propose a simulated retina visual structure to compress images and improve the robot efficiency. After separating the arm from the external environment, the robot establishes the model of the mechanical arm, and uses the "Minimal Resource Allocation Neural Network" to implement the robot's learning system. A developmental constraint implemented mechanism is applied to the robot system, so that, the robot adapts to the environment and completes the tasks in dynamic environment step by step. The experiments and simulations demonstrate that the robotic system, by imitating the process of the human development, gradually obtains the reaching ability.