Mobile Robot’s Sensorimotor Developmental Learning From Orientation and Curiosity

Abstract

Simulating biological intelligence has been proved to be an effective way to design intelligent robots, and simultaneously can solve the problems existing in machine learning methods. For creatures, their motor skills achieving is the first stage of learning. By combining two important cognitive elements: orientation and curiosity, this article proposes a new neurobiologically-inspired sensorimotor developmental learning method for the mobile robot. In this method, curiosity promotes robot's exploration of the environment, while orientation enhances robot's exploitation knowledge of the environment. The orientation cognitive algorithm is designed based on Skinner's operant conditioning theory, and its rationality is proved. The balance of exploration and exploitation, which is a key problem for all the cognitive learning method, is solved in this method. The developmental learning process can avoid fixed sensorimotor mapping space problem, and help reduce learning waste as well as computing waste. All of the developmental learning method's characters are finally verified via simulations on a virtual mobile robot.

Keywords

• Artificial curiosity
• autonomous robot
• developmental learning
• orientation
• sensorimotor skill