Adaptive Robust Cascade Force Control of 1-dof Joint Exoskeleton for Human Performance Augmentation

The control objective of exoskeleton for human performance augmentation is to minimize the human machine interaction force while carrying external loads and following human motion. This paper addresses the dynamics and the interaction force control of a 1-DOF hydraulically actuated joint exoskeleton. A spring with unknown stiffness is used to model the humanmachine interface. A cascade force control method is adopted with high-level controller generating the reference position command while low level controller doing motion tracking. Adaptive robust control(ARC) algorithm is developed for both two controllers to deal with the effect of parametric uncertainties and uncertain nonlinearities of the system. The proposed adaptive robust cascade force controller can achieve small human-machine interaction force and good robust performance to model uncer∗Address all correspondence to this author. †Zheng Chen was working at the Department of Mechanical Engineering, Dalhousie University, Halifax, NSB3H4R2, Canada, before May 2015, and starts to work at Ocean College, Zhejiang University, Hangzhou, 310027, China, in July 2015. tainty which have been validated by experiment. INTRODUCTION Exoskeleton for human performance augmentation is a human-machine interaction system which combines the power of the robot with the intelligence of the human. Due to the great potential in reducing the burden for people who carry heavy loads, it has attracted great interest from military and the industry. The development of exoskeleton began in the early 1960s and has experienced great progress in 21th century [1, 2]. The Berkeley Lower Extremity Exoskeleton(BLEEK) [3], Raytheon/Sarcos exoskeleton(XOS) [4]and the HAL exoskeleton [5–7] are typical examples. Exoskeleton robots can be classified according to the measurement method of the human-robot interaction: cHRIbased system and pHRI-based system [8]. The cHRI-based system,like HAL, measures the electric signals from the central nervous system to the musculoskeletal system of humans and use them as inputs to the robot controller. While, the pHRI-based 1 Copyright c © 2015 by ASME system, such as XOS and BLEEK, measures the force or position/velocity changes that are the results of the motions by the human musculoskeletal system and use them as inputs to the robot controller. The control objective of exoskeleton for human performance augmentation is to minimize the human machine interaction force while carrying external loads and following human motion. HAL used the EMG signal to compute the muscle torque and finally obtain the assistance torque for exoskeleton based on the muscle torque. The unreliability of EMG sensors limits the generalization of this control algorithm. BLEEK adopted sensitivity amplification control(SAC) in which a positive feedback controller using the inverse dynamics of the plant is used to increase the sensitivity to the force from the human. However, robust performance cannot be guaranteed due to the impossibility of obtaining an accurate inverse dynamics. The XOS system used human-machine interaction force as input of the robot controller. Unfortunately, few papers can be found on the detail of its control method. As for the human-machine interaction force control, several control algorithms have been proposed in the upper exoskeleton systems, such as admittance control in [9], impedance control in [10], and the human-machine cooperation controller design in [11]. Usually these exoskeleton controllers adopt a cascade control method in which the control system is grouped into two levels. High-level controller receives interaction force and produces reference commands(such as position, acceleration or force) while the low level controller is to track the reference from the high-level controller. It is observed that although there has been significant progress in the development of exoskeleton for human performance augmentation, a number of challenging issues still remain in the precise control of exoskeletons. First, the properties of hydraulic actuator(like significant oil compressibility, leakage) should be considered in the low-level controller design. Also, model uncertainties both from actuator and human-machine interface should be dealt with explicitly for good robust performance. Finally, the hardware physical constraint, low accuracy of the encoders in our system, should be considered when designing control algorithm. If the control algorithm requires velocity or even acceleration feedback, the system cannot be be implemented properly. In this paper, the modeling and cascade interaction force control of a 1-DOF hydraulically actuated joint exoskeleton is proposed. Springs with unknown stiffness are used to model the human-machine interaction. A simplified first order model for the hydraulic actuator is proposed which makes the low-level controller design simple. In the cascade interaction force control, the low-level controller is chosen to be a trajectory tracking controller since the force tracking of the cylinder is hard to achieved due to the oil leakage. The reference trajectory which can also be considered as human intent, is obtained form the high-level controller by making the integral of the human-machine interaction force become zero. The ARC control methodology is employed in both controllers to deal with the effect of parametric uncertainties and uncertain nonlinearities of the system. In order to guarantee the stability and performance of the whole system, a low-pass filter is used to describe the interaction effect between two subsystems. Comparative experiment verifies that the proposed adaptive robust cascade force control can minimize the human-machine interaction force and good robust performance to load change can be achieved. MODELING OF THE 1-DOF JOINT SYSTEM Full-Order Model The whole hydraulically actuated lower extremity is shown in Fig. 1. This paper only addresses the control problem of the knee joint exoskeleton. It includes three parts, as shown in Fig. 2, the human-machine interface due to the physical contact between the human and the machine, the mechanic system of 1-DOF joint(including the load), and the electro-hydraulic actuator. FIGURE 1. An exoskeleton system with electro-hydraulic actuation. For simplicity, a spring is used to model the the humanmachine interface and the stiffness of the spring is assumed unknown. The dynamics of the valve is neglected and the control input is assumed to be the displacement of the valve. The whole system is described by Eq.1, which includes the human-machine 2 Copyright c © 2015 by ASME