Font Size: a A A

Study On Reinforcement Learning And Its Application For Upper Limb Robot-assisted Rehabilitation Systems

Posted on:2016-08-20Degree:DoctorType:Dissertation
Country:ChinaCandidate:F C MengFull Text:PDF
GTID:1108330503953413Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
The upper-limb rehabilitation robotic system is one of the hot spots in the rehabilitation robotic field in recent years. But it often has complicated dynamic properties and uncertain working environments, and therefore cannot be adequately controlled by a conventional linear controller. To attack the problem, this thesis proposed a few control approaches for a class of upper- limb rehabilitation robotic systems based on the adaptive critic reinforcement learning algorithm. The main achievements are as follows:(1) An adaptive fuzzy actor- neural network critic-based controller(FANC) is proposed for rehabilitation robot. Utilizing the developed FANC controller, the rehabilitation robot can aid the patient achieve a semi- global asymptotic result even in the presence of uncertainties in the dynamic model and external disturbances. Simulation experiments show the effectiveness of the proposed control scheme.(2) A robust adaptive fuzzy critic algorithm(RFAC) is investigated for the upper- limb rehabilitation robotic systems with parametric uncertainty and external disturbances, the developed RFAC can utilize a robust fuzzy actor to generate an robust rehabilitation training control behavior under the guidance of a fuzzy critic and a robust integral of the sign of the error feedback technique. Lyapunov stability analysis shows the RFAC ca n satify the requirement of the patient and can yield a semi- global asymptotic result. Simulations verify the feasibility and effectiveness of the proposed control scheme.(3) To aid a person to achieve a desired tra ining trajectory with an optimal assistive force, a novel adaptive inverse optimal hybrid control(AHC) is proposed. The developed AHC consists of an adaptive critic controller and an inverse optimal controller. In the AHC, an adaptive critic controller is used to approximate the system model. The inverse optimal controller is appended to the adaptive critic controller to generate an optimal control input; which make the rehabilitation can supply the optimal assistive force for the patient. Finally, the AHC controller is proven(through a Lyapuno v-based stability analysis) to yield a semi- global asymptotic tracking. Simulation and experiment on a 2 DOF rehabilitation robot demonstrate the effectiveness of the proposed control scheme.(4) A new robust adaptive observer-based actor critic inverse optimal controller(i.e.,ROACI) is proposed for the robotic rehabilitation system without derivative states. The ROACI consists of a robust actor-critic-inverse optimal control system(RACI) and a robust dynamic adaptive observer. In the ROACI, The RACI is used to approximate the unknown rehabilitation system and generate optimal control inputs, whereas the robust dynamic adaptive observer is established to estimate the unmeasured derivative states. The developed approach(through Lyapunov stability analysis) is proved to achieve semi- global asymptotic tracking by selecting the suitable control gains. Simulation studies show that the improved approach can accomplish the desired training track task despite uncertainties in model parameters and external disturbance.(5) A new saturated robust adaptive inverse optimal switching control scheme(SRACI-SC) using multiple adaptive critics is proposed to deal with the constrained optimal control problem for the robotic rehabilitation systems with strong uncertainties. The developed SRACI-SC can generate the ideal control law for various training tasks. For repeating tasks, the developed control scheme is validity because of its ability to store the control actions for each training task with the multiple adaptive critics. Transient errors at the instant of task variation will be cancelled significantly when the memorized task occurs repeatedly in the training process. For a non-repeating new task, the proposed control scheme can utilize past experience to modify its control actions, which make the rehabilitation robot can achieve smooth robust optimal control performance through changing the suitable control rule. Lyapunov stability analysis shows the feasibility of the improved approach. Simulations and experimental studies are performed to show that the smooth robust optimal control performance can be accomplished by the prospoed control approach despite various uncertainties and disturbances in system.All the above- mentioned achievements effectively improve the control ability of the upper- limb rehabilitation robotic system and the practical application of adaptive critic-based reinforcement learning control alogrithm, and strengthern the autonomous control ability of the upper- limb rehabilitation robotic system within the adaptive critic-based reinforcement learnig framework.
Keywords/Search Tags:Upper-limb rehabilitation robotic system, Motion control, Adaptive critic, Inverse optimal control, Lyapunov stability analysis
PDF Full Text Request
Related items