Complementary Robot Learning of Actions and Concepts

Abstract

As expectations increase for robots to symbiosis with humans in real-world environments, the issue of robot knowledge acquisition is gaining attention. Since it is challenging to comprehensively design knowledge for robots operating in various environments, it is essential for robots to acquire knowledge autonomously. This study proposes a framework for complementary learning between actions and concepts by modeling curiosity, a universal desire to understand the environment. The experiments showed that the robot acquired the necessary reaching and grasping behaviors for concept learning and object concepts by integrating sensory-motor experiences of objects collected through these actions. Analysis of the acquired object concepts revealed that they reflect both the characteristics of the environment and the robot's behavior.