The Evolution Of Robot Behavior Based On Neural Network Integrated Approach

Traditional robot research was based on known structured environment, and researchers have accurate knowledge about robot and its environment. The main problems about this method are: (1) robot designers must have accurate knowledge about robot and its environment; (2) the hardware or software must be modified if robot and its environment need improving, this increased the burden of engineering implementation; (3) the adaptive ability of robot is weak because it must run under the plantation result from internal model of environment. Evolutionary robot is an important branch of robots science and technology more and more AX specialists pay attention to it because of its simplified structure and robust autonomous ability. Behaviorist believed that intelligent robot should be designed by using bottom-up method and should learn behaviors based on sensing-reactive by interacting with the environment. In this thesis, we present some new models and algorithms of robot behavior by using evolutionary neural network. The new model takes more emphasis on adaptive property of the robot, and let the robot run under the unknown environment. The robot can build the model of world by sensing environment and have self recovery ability. By learning continually and interacting with the environment, the robot can keep on adjusting the world and its own model and can finally run the environment. Our creative work include the following parts: 1. A new method for designing artificial neural network architectures, which is based on system and genetic algorithm, is presented in this paper. Production rules and parallel algorithm are used to solve traditional neural network design problems. The experiment results have proved that the algorithm can improve network performance and the speed of converge. 2. Obstacle avoidance, target approach, and wall following learning of evolutionary robot are realized by using artificial neural network in this paper. First, a robot learning environment and a robot model are presented and the implementation of evolutionary learning system is discussed. Then, the simulation experiments for basic behaviors and switch learning system of intelligent robot that adopted evolutionary learning mechanism are carried out. Finally, the simulation results are analyzed. 3. The behavior switch learning of evolutionary robot is realized by using artificial neural network in this paper. A robot learning environment and a robot model are presented and the implementation of evolutionary learning system is discussed. The simulation experiments 1^E are can-ied out for switch learning system of intelligent robot that adopted evolutionary learning mechanism. Finally, the simulation results are analyzed and the future research direction is given. 4. Obstacle avoidance, target approach, random and formation learning of evolutionary robot are realized by using artificial neural network in this paper. First, an advanced robot combined behavior learning method is presented and the implementation of evolutionary learning system is discussed by using neural network and AuRA architecture. Then, the simulation experiments are carried out for formation learning system of intelligent robot that adopted evolutionary learning mechanism. Finally, the simulation results of several type of formation type are given and analyzed. 5. This...