Path Planning and Human-Robot Interaction Control for Upper Limb Rehabilitation Exoskeleton
Abstract
Robot-assist rehabilitation constitutes a burgeoning field, supported by a corpus of empirical research positing the inherent efficacy of robotic systems. In general, the process for robot-assisted rehabilitation is to provide appropriate support to patients to assist them with regaining the performance of motor tasks. In comparison to end-effector type devices, exoskeleton-type apparatuses offer the advantage of exerting direct control over individual joints while concurrently mitigating deviant postures and motions. This affords a more precise and targeted regimen for the rehabilitation of each specific joint. In our laboratory, we conceived and fabricated the CLEVERarm—a six-degree-of-freedom exoskeleton. This innovative robot exhibits the capability to achieve precise alignment at the shoulder joint by orchestrating its joint movements in synchronization with the the human body. The development of this well-crafted device now positions us on the cusp
of achieving a comprehensive and optimally functioning exoskeleton system. The ensuing focus of our research centers on the refinement and establishment of sophisticated control algorithms, constituting the principal subject of investigation in this study.
In the realm of robot-assisted rehabilitation, a frequently encountered undertaking pertains to the pursuit of trajectory following. This task encompasses the formulation of a predetermined trajectory, the conception of a control algorithm, and the evaluation of performance. To tackle these three distinct domains, this study has introduced dedicated solutions for each of them. First, with regards to trajectory tracking, the study introduces a novel trajectory planning methodology, yielding
more anthropomorphic reaching motions. The presented work specifically focuses on solving the resolution of redundancy in reaching motions. This method is based on the notion of maximum manipulability as defined in Kim et al, 2012 with two additional refinements by the current authors: (1) Swivel angle primitive and (2) Combined straight line primitive. A human-robot configuration transformation model is also designed for mapping the coordinates from the human upper-limb
configuration space to the exoskeleton joint space. In the context of designing control algorithms, this investigation formulates two controllers. The first controller serves as a real-time control algorithm, responsible for furnishing appropriate torque inputs to individual robot joints. It relies on a human-robot interaction index denoted as J. The second controller functions as an iterative trajectory refinement approach during each training iteration. It incorporates a power transform function meticulously designed to modify conventional symmetric bell-shaped velocity profiles. The realtime assessment of performance, as quantified by the index J, is attained through the utilization of fuzzy logic techniques. Furthermore, the evaluation of performance across successive training iterations is accomplished through the introduction of a criterion embedded within the velocity profiles. Surface electromyography (sEMG) signals, owing to their considerable predictive capacity with regard to upper limb motion, hold promise for prospective applications in robot-assisted rehabilitation. In this context, our study introduces a deep learning model tailored to forecast upper limb motions by leveraging the information encoded in the sEMG signals. Within the scope of this investigation, two distinct models are formulated. The first model pertains to real-time intention prediction within the upper limb’s configuration space, while the second model addresses the prediction of entire trajectories for upper limb reaching motions. To enhance data pre-processing, a personalized standardization (PSD) technique is introduced, which yields notably superior accuracy compared to conventional data pre-processing methodologies. Experimental verification results are presented for the aforementioned algorithms, demonstrating their efficacy and suitability for application in the field of robot-assisted rehabilitation.
Collections
Citation
Nie, Kuang (2023). Path Planning and Human-Robot Interaction Control for Upper Limb Rehabilitation Exoskeleton. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /202973.