A Learning-Based Hierarchical Control Scheme for an Exoskeleton Robot in Human–Robot Cooperative Manipulation

Abstract

Exoskeleton robots can assist humans to perform activities of daily living with little effort. In this paper, a hierarchical control scheme is presented which enables an exoskeleton robot to achieve cooperative manipulation with humans. The control scheme consists of two layers. In low-level control of the upper limb exoskeleton robot, an admittance control scheme with an asymmetric barrier Lyapunov function-based adaptive neural network controller is proposed to enable the robot to be back drivable. In order to achieve high-level interaction, a strategy for learning human skills from demonstration is proposed by utilizing Gaussian mixture models, which consists of the learning and reproduction phase. During the learning phase, the robot observes and learns how a demonstrator performs a specific impedance-based task successfully, and in the reproduction phase, the robot can provide the subjects with just enough assistance by extracting human skills from demonstrations to prevent the motion of the robot end-effector deviating far from desired ones, due to variation in the interaction force caused by environmental disturbances. Experimental results of two different tasks show that the proposed control scheme can provide human subjects with assistance as needed during cooperative manipulation.

Learning for a Robot: Deep Reinforcement Learning, Imitation Learning, Transfer Learning

“…Methods of robot learning address the lack of accurate object models and dynamic changes in complex environments. The learning process is also simplified by visually extracting information from expert presentations [134].…”

Adaptive Reinforcement Learning-Enhanced Motion/Force Control Strategy for Multirobot Systems

“…According to Lyapunov's direct technique, authors in [5] presented the robust controller for a bilateral teleoperation under time-varying delays and without relative motion. Recently, the backstepping technique of networked robotics has achieved much attention [3,11,21]. In [3], a backstepping method is utilized to implement an adaptive trajectory tracking control design for a tractor trailer, where the proposed control scheme can stabilize for cascade control systems.…”

“…In [3], a backstepping method is utilized to implement an adaptive trajectory tracking control design for a tractor trailer, where the proposed control scheme can stabilize for cascade control systems. For the purpose of easily computing the backstepping method and handling the full-state constraint, authors in [21,22] pointed out the appropriate term α in the asymmetric barrier Lyapunov function (ABLF). is backstepping method is extended in [11] to deal with unknown dead zone nonlinearity, time-varying term using the inputdriven filter term and Nussbaum function, respectively.…”

Detection of Small Target Using Schatten 1/2 Quasi-Norm Regularization with Reweighted Sparse Enhancement in Complex Infrared Scenes

“…The efficient and robust performance of detection has an important role to play in these applications. However, small targets may be buried in complex infrared scenes with low signal-to-clutter ratios deriving from high bright noise and strong thermal radiation clutters [3]. And they tend to be weak and/or even negligibly small without concrete shape and discriminating textures owing to a long distance between projected targets and imaging sensor [4].…”

“…Representative single target images from the datasets and the separated target images obtained by six low-rank recovery-based methods (1)(2)(3)(4). are four representative single target images from the tested datasets.…”