This dissertation focuses on the research of soft computing on sensor-based autonomous navigation of an omni-directional mobile manipulator with behavior-based control. Firstly, relevant aspects of mobile robots, which is a family of intelligent robots, are introduced. The research status of intelligent control on architecture of mobile robot and soft computing on sensor-based autonomous navigation of mobile robots are reviewed respectively. Also, the background and roadmap of this dissertation are presented. Secondly, the frame of control system for the omni-directional mobile manipulator is established, which is based on module feature. Then a new behavior-based hybrid architecture and key points of the design for behavior coordination mechanism are presented. The validity is described by two illustrative examples. Thirdly, considering the usual structure of neural-fuzzy inference network, node function and inference mechanism, this dissertation proposes an extended neural-fuzzy inference network, which can overcome the imprecise reasoning and the loss of information. To solve the problem of training data based learning for the extended neural-fuzzy inference network, an on-line and an offline learning algorithm based on mamdani fuzzy model and a learning algorithm based on mamdani fuzzy model with certainty grades are presented respectively, which consist of three leaning phases. Simulation results of fuzzy identification for complex systems show effectiveness of the structure of extended neural-fuzzy inference network and learning algorithm, which is also applied successfully to the development of navigation controller for wall-following of mobile platform. Fourthly, to solve the problem of non-training data based learning for the extended neural-fuzzy inference network, a class of reinforcement learning based neural-fuzzy control system is established. This dissertation designs a learning algorithm for the reinforcement learning based neural-fuzzy control system from another point of view, which is trying to convert reinforcement learning problem into training data based learning problem. Also a reinforcement learning based radial-basis function network control system is established. Simulation results of obstacle-avoidance for the mobile platform in unknown environments show the validity of the structure and learning algorithm of the reinforcement learning based control. Fifthly, three navigation simulations based on composite behavior are designed, which verify the effectiveness and applicability of the presented architecture, the structure and learning algorithm of the neural-fuzzy control system for sensor-based autonomous navigation of mobile robot in unknown environments. Finally, the obtained research results are summarized and future work is addressed.
修改评论