Study On Robot Learning By Imitation

The ultimate goal of intelligent robotics research is to make machines behave intelligently, that is to learn from previous experience and adapt to the ever-changing environment. A robot estimates the environment changes through sensing, and responds to these changes through actions. The aim of robot control is to construct an appropriate structure between sensing and action to perform a task.A burgeoning strategy for robot control behaviorism, which abandons traditional AI and gets rid of some drawbacks of traditional AI, reduces the requirement for environmental recognition while simplifies the design process for designers. As a result, behaviorism becomes an issue in robotic researching field, and behaviorism is the direction for future robotic research. However, new problems need to be resolved for the new strategy, i.e., how to create the sensory-motor description for the robot. At the beginning of the dissertation two problems needing to be answered for robot control are shown. And then, combined with the behaviorism, a new robot learning method, which is robot learning by imitation, is discussed. The promise of this learning strategy is that it provides robots with an efficient non-verbal means of communication. Because it is non-verbal, it does not require the teacher and the robot to 'speak the same language'. The robot can learn from other robots that are of different species or are built from different hardware. This means that the sensory-motor description can be learnt by imitating a demonstrator's behavior. At the last of this dissertation, a research using Pioneer3DX platform with 5D0F robotic arm to imitate a demonstrator's behavior is performed. Using the on-board vision tracking system to observe the demonstrator's behavior, then obtaining the movement information, finally control the robotic arm to imitate the demonstrator's movement.