| Modular and reconfigurable robot (MRR) consists of standard robot modules andelectromechanical interfaces that can be assembled to desirable configuration forsatisfying the requirements of various tasks. With the concept of modularization, therobot modules incorporate the units of communications, power electronics, controlsystems, sensors and actuators, so that the configuration of the robot can be changedaccording to the requirements of the tasks under different external environment and theconstraints, and make the robot to adapt to the new working environment afterreconfiguration. Generally, the number of configurations of MRRs can be generateddepending on the types and degrees of the joint and link modules as well as the number ofthe interfaces, etc.. Compared with the conventional robots, an MRR is with theadvantages of achieving many kinds of assembly configurations and provide differentoutput powers, e.g., achieving structural flexibility of the robot by reconfiguration, addingand subtracting the modules; providing a cost-effective test platform for the developmentof new robot products and promoting the developments of the new technologies; shortingthe development cycle of the new technologies and reducing the costs of the development,testing and maintenance of the robot in the long term. At the same time, an MRR needs anappropriate control system to provide accuracy and stability after reconfiguration.When an MRR is performing a task, the contact and constraint take place on theoutside environment is inevitable. In order to compensate the constrained force of therobot joints and to obtain a precise control, the study on joint torque feedback,compensation and dynamic control of the time varying constrained MRR is an importantproject in the research field of MRR. Until now, numerous studies have been carried outon the joint constrained force compensation and control method based on direct torquesensing technology. However, installing joint torque sensor may reduce the reliability androbustness of the robot and complicating the structure of the module. Therefore, it is animportant issue to be addressed that studying the joint torque estimation andcompensation control method for time varying constrained MRR without usingforce/torque sensor. For the purpose of ensuring the stability and accuracy of the MRRafter reconfiguration, and completing the tasks under the time varying constraint, there isanother important project in which studying the dynamic control method for the timevarying constrained MRR. With this reason, the compatibility and reconfigurability of thecontrol system should be considered and the designed controller should be with a goodperformance for different robot configuration. In order to meet the requirements above,the conventional controller need to consume large amounts of computing resources, sothat it is hard to ensure the stability and reliability of the controller when the robot system structure is too complicated. Therefore, it is necessary to study the dynamic controlmethod for an MRR with modular design idea, low complexity, fast operation speed andstrong identification and compensation ability for the model uncertainty.This paper addresses the problems of joint torque estimation and decentralizedcontrol for the time varying constrained MRR, and mainly studies on the dynamicmodeling method of the time varying constrained MRR, the joint torque estimationmethod based on the harmonic drive model, decentralized reinforcement learning optimalcontrol method based on Actor-Critic-Identifier, decentralized sliding mode controlmethod based on variable gain super twisting algorithm, decentralized integral slidingmode control method based on joint torque estimation, etc.. And this thesis is organized asfollows in which the main research works are included.1. Introduce the background and significance of this thesis, and then overview theresearch status and hot issues of MRR.2. Based on Newton-Euler iterative algorithm, the dynamic model of MRR isformulated, according to a set of effective forward and reverse iteration equation andconsidering the force/torque which enforces on the joints and links and the coupledforce/torque among the joints. Based on local joint dynamic information, the dynamicmodel of the robot is described as a synthesis of interconnected subsystems, in which thedynamic characteristic is analyzed, then the dynamic model of the subsystem of the timevarying constrained MRR is formulated. According to the double external force actedspring mass system characteristic, the harmonic drive gear which is embedded in the jointmodule of the MRR is studied and a joint torque estimation method based on harmonicdrive model is proposed.3. Based on the Actor-Critic-Identifier scheme, a decentralized reinforcementlearning robust optimal tracking control method for time varying constrained MRR isproposed in the condition of existing coupling model uncertainty. The critic-NN andactor-NN are used to approximate the optimal Q-function and the optimal control policy,as well as the identifier is adopted to identify the model uncertainty. Then, thedecentralized reinforcement learning robust optimal tracking controller is designed andthe optimal conditions under the HJB equation is satified. Finally, the stabilities of theIdentifier and the controller are confirmed by the Lyapunov theory.4. Study on the decentralized sliding mode control method for MRR based onVGSTA. First, a decentralized terminal sliding mode control method is proposed based onVGSTA-ESO, and the genetic simulated annealing algorithm is used to adaptively adjustthe determined parameters of the observer. This method combines the estimation andresist disturbance abilities of the ESO and the system error convergence ability of theVGSTA, so that the desired trajectory can be tracked within a finite time interval, and theestimation error can converge to zero in finite time. Second, a novel decentralized integralnested sliding mode control approach for a time varying constrained MRR is proposed. Combining the concepts of integral sliding mode with nested sliding mode, an integralnested sliding surface, which is designed by employing the pseudo-sliding surface blockand the sigmoid function, is implemented. Then, the decentralized integral nested slidingmode controller is designed to compensate the model uncertainty and to reduce thechattering effect.5. Study on the decentralized integral sliding mode control method for harmonicdrive transmission based MRR under the free space and time varying constraint. First,based on the joint torque estimation method which is proposed in section2, the nonlinearvelocity estimator and the decentralized integral sliding mode controller are proposed thatusing only position measurements, then experiments are performed for a3-DOF MRR tostudy the effectiveness of the proposed method. Second, a decentralized integral slidingmode control method for a time varying constrained MRR is proposed, and addresses theproblems of robot trajectory tracking in the condition of existing coupling modeluncertainty and time varying constrained force. Combining the integral sliding modedesign with the decentralized control scheme, the decentralized controller, which is usingonly local joint dynamic information, is designed to compensate the model uncertaintyterms that including the friction modeling error, harmonic drive torque ripple and thedynamic coupling, and the chattering effect of the controller is reduced. Therefore, thehigh-performance trajectory tracking control for both the robot joint and the end-effectorcan be achieved.Finally, the conclusions and the perspectives of future research are presented at theend of this thesis. |
PDF Full Text Request