Research On Control Of Pneumatic Artificial Muscle-actuated Robots

With the rapid development of the artificial intelligence technology,there is an increasing demand for robots with higher degree of autonomy and better human-machine comfort.The development of robots driven by large-size and inflexible traditional actuators(e.g.,motors,hydraulic actuators,cylinders,etc.)is gradually falling into a bottleneck.In order to adapt to the new era of intelligent revolution,safer and more flexible pneumatic artificial muscles(PAMs),also known as pneumatic muscle actuators,are increasingly becoming the core of intelligent,interactive,and flexible robots,which can replicate the features of natural biological muscles with specific material mechanisms.Robots driven by PAMs have extremely high mobility and amazing power density,which can flexibly and closely contact with humans and environment,and can also improve the compliance and safety of entire systems,so as to achieve comfortable and friendly human-machine interactive control.However,while robots benefit from good compliant characteristics by PAMs,they also suffer from a series of complicated characteristics induced by PAMs,e.g.,hysteresis,nonlinear characteristics,low response frequencies,etc.These potential control problems are unfavorable to accurate modeling and motion control,which will badly limit their wide applications.To this day,there are a certain amount of researches on kinematics/dynamics modeling works and positioning/tracking control methods for robots actuated by PAMs.However,from the perspective of practical applications,there are still some unsolved problems requiring breakthroughs as follows: 1)The problems about unidirectional input constraints of PAM-actuated robots are not discussed in existing references.The chattering problem is aggravated easily by the violent use of discontinuous sliding mode control laws to deal with system uncertainties.2)In actual working conditions,the system may face unexpected effects of continuous unknown disturbances.However,for common active disturbance rejection control methods,differential operations of extended state observers(ESOs)are frequent,and the parameters are difficult to choose,which may cause high gains and spike problems.3)There are skid and friction problems when the joint is driven by the antagonistic action with multiple PAMs.Most of relevant works have not introduced the vibration damping mechanism(e.g.,springs)to eliminate residual oscillations.Moreover,most of existing models are equations about torques,but torques are not the real control inputs of PAM-actuated robots.4)Most existing methods do not limit system overshoots,and rarely discuss the singularity of the control law,which may lead to safety risks.5)Due to the limited driving capacity of actual actuators,input constraints,such as saturations and dead zones should not be ignored.In addition,for the unknown terms in the actual system are usually difficult to satisfy the linear parameterization(LP)conditions,which makes common adaptive methods have limitations in dealing with the above problems.Hence,the convergence rate of the error is reduced.For the above challenges,for both theoretical and practical considerations,precise modeling and high-performance tracking control of PAM-actuated robots(including 2-link PAM-actuated robots)are studied in depth.The main contributions of this thesis are listed as follows:1)Continuous adaptive robust control for PAM-actuated robots with unidirectional constraints.Considering existing issues of PAM-actuated robots(e.g.,unidirectional constraints,parameter uncertainties,and so on),a continuous robust control scheme and an adaptive update law are proposed,which can compensate for uncertainties,suppress external disturbances,and satisfy unidirectional constraints,so as to realize accurate positioning and tracking control of reference trajectories.Compared with traditional sliding mode control methods,the proposed method is continuous,which can avoid the chattering problem.Without any linearization operation,the closed-loop system is theoretically proven to be asymptotically stable at the equilibrium point.In addition,an existing adaptive control method is chosen for comparison,a series of hardware experiments are carried out to verify the good control performance and robustness of the proposed method.2)Disturbance estimation-based nonlinear control for PAM-actuated robots.In the presence of continuous unknown disturbances,a nonlinear control method based on disturbance estimation is designed for PAM-actuated robots,which can estimate and compensate for the disturbances online,and achieve accurate tracking control simultaneously.First,system uncertainties,unmodeled dynamics,and external disturbances are expressed as a lumped disturbance term by model transformation,which can be accurately estimated by the sliding-window iterative integral operations and the least squares algorithm.Then,a sliding mode control law is employed to eliminate estimation errors,and ensures asymptotic stability.Based on accurate estimation results,the pressure of sliding mode control can be alleviated,and the chattering problem is greatly avoided.In addition,a detailed stability analysis is given,and the effectiveness and robustness of the proposed method are verified by hardware experiments under the persistent input disturbances.3)Dynamic modeling and motion control for 2-link PAM-actuated robots.A new 2-link PAM-actuated robot is designed and built,which can eliminate residual oscillations in time.Further,the configurations of hardware/software platforms are given in detail.Then,an accurate dynamic modeling method is proposed for this platform.By taking the derivative of Lagrange's function about muscle lengths,and introducing the three-element model of PAMs,a dynamic model for air pressure is established,which is more intuitive and effective than existing torque-based models.4)Energy-based nonlinear control for 2-link PAM-actuated robots.Based on the derived dynamics,an energy-based nonlinear control method is proposed,which can satisfy preset constraints of system overshoots and special coupling terms,and simultaneously ensure accurate positioning/tracking performance.On the basis of system energy analysis,a new energy storage function is designed,which lays the foundation to construct the Lyapunov function candidate.The designed auxiliary functions ensure that system overshoots and special coupling terms are always limited in safety ranges,which can improve system safety and reduce energy costs.Both controller design and stability analysis are carried out on the original nonlinear model,and the tracking errors can be proven to be asymptotically convergent to zero without linearization operations.Hardware experimental results of the proposed and the comparative methods are presented in this thesis.5)Adaptive fuzzy-sliding mode control for 2-link PAM-actuated robots with multiple input constraints(e.g.,unidirectional inputs,saturations,dead zones,etc.).For the limited actuating abilities of actual components,an adaptive fuzzy sliding mode control scheme is presented without a priori model structures and parameters,which can realize complicated disturbance rejection and simultaneously accurate trajectory tracking.The proposed method not only considers nonlinear input constraints and the estimation of complicated disturbances,but also ensures finite-time convergence of tracking errors.The proposed fuzzy update law can estimate input constraints,parameter uncertainties,unmodeled dynamics,and external disturbances in real time,and the approximation error vector can be further compensated for by the presented sliding mode control scheme.In addition,the finite-time convergence of the tracking error vector is proven through the closed-loop stability analysis.In order to validate the actual control performance of the proposed method,abundant comparative experimental results are provided in this thesis.