Programming Robots to Learn Simple Tasks Autonomously (Project I)

Abstract

Scholars found that intrinsic motivation like curiosity of human is crucial especially during their childhood stage development; they seem to have intrinsic reward for their exploratory activities. To equip robot with the most sophisticated learning mechanism that is similar to human and be able to adapt to complicated environment become the most challenging problem nowadays. In this project, an artificial intelligent system called "Intelligent Adaptive Curiosity" (IAC) is developed through modelling curious behaviour of human. A real robot equipped with sensors and motors has been used. When it is put in novel environment, it prioritize learning activities autonomously according to levels of difficulty of each action. Each action is associated with an individual neural network. The neural network calculates the learning progress and trained with sensors and motors responses. The results indicate that IAC system motivates the robot to learn in novel situation. The motivation for the robot to learn simulates human curiosity, so new task can be learned. It increases the difficulty of learning task progressively according to the progress of its learning. Also the neural network with ability of generalisation is suitable for learning actions in this project. The result suggests that modelling curiosity is crucial for the robot to learn new skills progressively.