Study On Robot Gesture Imitatiomn Based On Vision

Currently, The research of robotics is developing to intellectualization and versatility. In order to meet the requirements of many kinds of application environment, the robot needs to be more flexible, as well as certain learning capability. But the traditional methods of constructing robot only aim at specific duty, also need massive manual programming process. it's a hard work and difficult to satisfy such request. Therefore, many researchers are actively exploring the new methods to construct intellectual robots, for the liberation of humanity from the complex programming process, and those methods ensure that they has more energy to be engaged in the creative work.Compared with other kinds of methods used to construct intelligent robot, the method of imitation learning has received more and more attention because of its advantages of more flexible, higher efficiency, and so on. Imitation learning has avoided the complex programming process aims at specific duty, enables the interaction between robot and environment through robot's own sensor system. The robot can learn new knowledge and methods to solve problems through the interaction with environment. The imitation enable the robot individual has a higher flexibility. By observing demonstrator's movements, the robot can study the useful knowledge fastly, that let the robot adapt the new environment rapidly. And, compared with the traditional robot learning process as an individual, the imitation learning greatly enhanced the learning efficiency. Many experiments has proved that, the imitation learning is an effective robot learning method.This thesis has done an detailed introduction about the research approach and current research situation of " Robot Gesture Imitation ", and has set up an simulation system for robot imitation learning, completed the imitation of instructor's arm movements. Inspired by "Autonomous mental development theory", this article tries to put forward a method of setting-up the "sensor - action " mapping based on HDR method and neural network. To a certain extent, this method has solved the problem that the network converges slowly when the input variable is complicated vectors, and has succeed in applying this method to robot gesture imitation. The main work of the thesis is as follows:1,Analyzed the advantages of robot imitation-learning and its current research situation;2,Extracted the instructor's angles of shoulder joint and elbow joint through the pattern process method;3,Inspired by "Autonomous mental development theory ", has completed the prediction of joint angle by the method of HDR combined with neural network, and analyzed the results with methods of error and relevancy analysis;4,Set up simulated arm model, used the angle of joint to control the fictitious arm's movement, and has completed the recurrence of instructor's movements.