Study On The Kinematics, Dynamics And Control Method Of Reconfigurable Modular Robots

Modular and reconfigurable robot consists of link modules and joint moduleswith standard interface. These modules can be quickly assembled into robots indifferent configuration that has different kinematics and dynamics to adapt todiverse work and environment. Compared with traditional fixed robots, modularand reconfigurable robot can fit the working requirement in large range, and alsotake advantages of low costs, convenient maintenance, and portability and so on.Thus, they enlarge the application space in rapid manufacture, military affairs,aerospace, and industry. In this dissertation, the contribution is theoretical researchon kinematics automatic generation, dynamics automatic modeling and controllerautomatic design for modular and reconfigurable robot.First of all, module unit design is an important issue in modular andreconfigurable robot because many kinds of robot configurations in differentgeometric forms are assembled through changing the order of modules. As muchmore kinds of modules are made, the robot structure will be more complicated;moreover, automatic solutions of kinematics, dynamics and configurationoptimization will be more difficult. Hence, in order to simply the system structure,two link modules and two joint modules concept models are designed in this paper.Link rules among modules are constituted and configuration frames of modulesare established in terms of avoiding of redundancy. Based on graph theory,common kenematic graphs in mechanics design are used to describe theconfiguration of modular and reconfigurable robot.However, the kinematics also is significant in robot research, which focuseson the geometrically kinematical relation among the parts of the robot. There aremany solving methods on fixed robot kinematics. But because of many differentconfigurations of modular and reconfigurable robot their kinematics equation can'tbe solved one by one. In this paper, kinematics frames of modules are set up basedon configuration coordinate description, database containing geometric parametersof modules is built, and kinematics is automatically generated on the basis of localproduct-of-exponent formula and kinematic graphs. On the aspects of solving thereverse kinematics, there are two methods, which are analytical method andnumeric method. The former works out the solution of inverse kinematics througha set of equations with respect to joint variables. Its superiority is that it can solveout the more efficient close solution. However, analytical method in existence lieson kinematics structure of robot. It is very difficult for analytical method to solvethe close solution of inverse kinematics. In this paper, Newton-Raphson numericmethod is used to solve the inverse kinematics of modular and reconfigurablerobot, and singularity robust inverse concept is applied to make robot pass thesingularity position.Then, dynamics of robot describes the relation between motion and force ortorque exerted on joints. Dynamics model plays an important role in mechanicaldesign, control algorithm, motion simulation and forward feed compensation. It isdivided into two parts: forward and inverse kinematics. Up to now, dynamicsequation in geometric form is up to date. This Lie Group based method, considersSpecial Euclidean Group as Lie Group, escaping from the special concept intraditional dynamics algorithm. Based on the kinematics model, a group ofiterative Newton-Euler equations is derived through generalized velocity,acceleration and force. These equations destroy the model structure of dynamics,which is convenient to design the controller, thus, it turns hard to get compositivecontrol law. In this paper, global matrix description about dynamics equation isproposed.Finally, control algorithm design is more challenge field on modular andreconfigurable robot. Why complicated is that the controller must guarantee thereliability in spite of every configuration. In past, many methods were presented,such as computed torque control, adaptive control, fuzzy control, sliding control,robust control and so on. Some of them require accurate model of dynamics, somerequire model with no unmodeled dynamics, and others requires the bound ofuncertainty exist. All and all, these restrain applications of these control algorithmsin actual robot control. Recently, due to the potential of approximating arbitrarynonlinear mapping, neural networks have been popular in robot control. Twothoughts appear, one is using neural networks to approximate the whole dynamicsmodel;the other is to estimate and compensate the uncertainty. Here, whole frameof controller is proposed based on automatic model technology of modular andreconfigurable robot. Considering information of dynamics, model uncertainty andunmodeled dynamics, neural networks compensating controller based oncomputed torque control is proposed. Compensator adopts function link neuralnetworks, and a robust term is added. Under the stability prove through Lyapunovtheory, the tracking performance is enhanced and reliability is proved fromsimulations conducted on modular and reconfigurable robot in two configurations.