Study On Decentralized Adaptive Backstepping Control For Reconfigurable And Modular Robots

With the progressive development of robot technology continue to expand its application areas, it is not only widely used in traditional industrial production and manufacturing, but also in the aerospace, deep-sea exploration, dangerous or adverse conditions, also received the special environment of a large number of applications. In a particular environment, the fixed configuration of the robot-specific tasks required to design, by changing the parameters of the robot can perform different tasks. building a lunar base station, in order to design a fixed configuration of the robot system to perform multiple operations is very difficult or impossible task. However, in many unknown or changing environment, such as nuclear reactor power stations, polar research station or In such circumstances, reconfigurable modular robot to show the traditional fixed parameters do not have the advantage of the robot, it can re-form by changing the parameters of the robot to adapt to different requirements of different tasks. Although reconfigurable modular robot forward and inverse kinematics, dynamics automatically generated, centralized control and distributed control of research has gone deep, but there are still some challenging issues and areas of further research to be, such as inverse kinematics the model handles the cross-linking after the decomposition, vibration control, fault identification and fault-tolerant control of the research is not mature enough, so the reconfigurable modular robot for further research has great theoretical and practical significance.In this dissertation, the main research field of the reconfigurable modular robot is in detail analyzed and at the same time work out the reconfigurable modular robot dynamic model using decentralized adaptive neural network control and decentralized adaptive iterative learning control method. The essence of this dissertation is listed as below:First, base on rigid body Newton-Euler iterative algorithm, a iterative Newton-Euler function of the system is built by iterative of generalized velocity generalized acceleration and inverted iterative of generalized force. The dynamics function of reconfigurable and modular robot is set up. Consideration of each joint modules of reconfigurable robots into a subsystem, by separating terms depending only on local variables from those terms of other joint variables, given the dynamic model of the sub-system.Second, decentralized adaptive neural network control scheme is proposed for the dynamic control for trajectory tracking problem of reconfigurable and modular robots. Considering the sub-system dynamics model, in the case of kinetic model parameters assuming known, general backstepping techniques to design the ideal decentralized control law. Neural networks are used to approximate the unknown function in the deal decentralized controller, compensate interconnections and the effects of approximation errors, adaptive laws are designed to adjust the weights of the neural network. All adaptive algorithms in the subsystem controller are derived from the sense of Lyapunov stability analysis, so that the resulting closed-loop system is stable and the trajectory tracking performance is guaranteed. Considering reconfigurable robot joint velocities are unpredictable, adaptive state observers are designed, based on the state observers, using backstepping techniques designed decentralized controller. Neural networks are used to approximate the unknown function in the deal decentralized controller, all adaptive algorithms in the subsystem controller are derived from the sense of Lyapunov stability analysis, so that the resulting closed-loop system is stable and the trajectory tracking performance is guaranteed. Two degrees of freedom for the two different configurations of reconfigurable robots show that the proposed two control algorithm, feasible.Third, decentralized adaptive iterative learning control scheme is proposed for the sub-system dynamic function of reconfigurable and modular robots. This approach is using auxiliary filtered error function, after mathematical derivation got together the unknown functions and interconnections, and used to approximate by the neural network, greatly reduce the complexity of the controller design. Considering the sub-system dynamics model of the reconfigurable robot, assuming the interconnection are bounded, the proposed decentralized adaptive iterative learning control algorithm can guarantee the closed loop system stable, and error convergence in the iteration domain. Neural networks are used to approximate the unknown function and interconnection, adaptive laws are designed to adjust the weights of the neural network. All adaptive algorithms in the subsystem controller are derived from the sense of Lyapunov stability analysis, so that the resulting closed-loop system is stable and the trajectory tracking performance is guaranteed. This approach makes the system contains only one unknown term, not two, so little use of a neural network system, the control structure than the decentralized adaptive neural network control method is more simple. Through the simulation, decentralized adaptive iterative learning control method can achieve effective control of the system, but also in the simulation results than the decentralized adaptive neural network control method is more accurate.Finally, the author of the text is summarized, and on this basis, combined with its own research experience, and further research were discussed.