Towards autonomous ergonomic upper-limb exoskeletons: A computational approach for planning a human-like path

Highlights

• A computational approach for path generation in upper-limb exoskeletons.

• A novel framework for generating human-like paths in the upper-limb exoskeletons.

• Path generation considering the scapulohumeral rhythm.

• An analytic transformation for mapping paths between the human-arm and exoskeleton configuration spaces.

Abstract

Computational path planning approaches can enable development of autonomous rehabilitation and assistive exoskeletons. Using a human-like reference behavior for such wearable systems can ensure safe, effective, and intuitive human–robot interaction. This is of significant importance since the quality of interaction and ergonomic considerations have a substantial effect on technology usability and acceptance by the users. This paper proposes a novel framework for generating human-like paths for wearable exoskeletons in the shoulder-elbow level. The introduced method is a two-stage process where a human-like reference path is planned in the configuration space of the human arm, followed by an analytical transformation that directly maps the derived path to the configuration space of the exoskeleton. The analytical mapping presented is a function of the kinematic parameters of the system and can be adapted for other upper-limb exoskeletons. As a case study, the proposed method is used for generating human-like reference motions for a six-degree-of-freedom exoskeleton supporting scapulohumeral rhythm, glenohumeral rotations, and elbow flexion/extension. Firstly, it is shown that reaching motions associated with activities of daily living can be predicted with high accuracy in the human joint space. This is demonstrated by analyzing the experimental data collected from healthy subjects. Subsequently, it is verified through kinematic analysis that the transformation of generated paths to the exoskeleton configuration space does not alter their spatial profile in the task space.

Introduction

Upper-limb exoskeletons have proven to be a promising technology for assistive purposes, occupational applications, and rehabilitation of motor impairments [1], [2]. In all these applications, achieving an intuitive and ergonomic interaction between the exoskeleton and the user is a critical criterion for usability, acceptance, and potentially the effectiveness of the system [3], [4], [5]. A core requirement for realizing such an interaction is the capability of the system in producing natural and human-like movements [6]. This is especially important for therapeutic exoskeletons where the robot has a more dominant role in controlling and correcting the motions of the limb based on a prescribed reference behavior [7], [8]. Currently, therapists are responsible for devising the training motions, however, the effectiveness and autonomy of a robot-based therapy can be significantly improved by utilizing a human-like reference motion generation algorithm which can be adjusted based on individual patients’ conditions (e.g. range of motion, spasticity level, etc.). It can be argued that such a feature can be very useful for assistive systems as well [9], [10]. Human-like motion generation along with high-level intention detection algorithms are essential for achieving autonomous assistive technologies that can be integrated into the daily lives of the disabled.

Various methods have been used in the literature for reference motion generation of exoskeletons. Using the recorded motion of the affected limb moved by a therapist is a common approach. In this method, the motion of a healthy individual is recorded by a motion capture system [11], [12], or by an exoskeleton attached to the patient and back-driven by the therapist [13]. In a similar approach, mainly used in bilateral systems, the motion of the healthy side of the patient can be mirrored on the affected side [14], [15]. These methods necessitate the presence of a therapist to move the patient’s arm to generate every motion or require additional equipment such as bilateral exoskeletons, motion capture systems, and a technician familiar with the equipment. These restrictions limit the application of exoskeletons to clinical settings rather than potential home-based options and increase the cost and complexity of the systems. To address the aforementioned limitations and to automate the path generation process, computational approaches could be used [16]. Computational models are grounded on theoretical studies on motion generation principles of the human Central Nervous System (CNS). The majority of these computational models use an optimization-based approach in accordance with the hypotheses that CNS plans the upper limb motions such that a specific cost is minimized. Motion jerk in task space [17], [18], [19], angular jerk [20], squared change of the joint torques [21], [22], work done by joint torques [23], peak work [24], and energy [25] are examples of the cost functions studied in the literature.

Among the diverse sets of computational models, Minimum Jerk (MJ) in task space has been widely adopted for motion planning of upper-limb exoskeletons [17], [18], [26]. This can be partially attributed to the computational ease of planning motions in task space without the need to consider the specific kinematic and dynamic properties of the actual robot. However, due to the redundant nature of human arm kinematics, utilizing a task-space path planning strategy necessitates use of an additional algorithm for resolving redundancies. Another model used in the literature for exoskeleton motion planning is based on minimizing the squared derivative of joint trajectories [27]. This choice of the cost function results in minimum distance paths in the configuration space (MDC) and is therefore computationally very efficient. It should be noted that an MDC path in the configuration space (c-space) of an exoskeleton is not necessarily equivalent to the MDC path in the configuration space of the human arm model (except for exoskeletons whose shoulder joint axes align with the biologic axes of the human shoulder [28]). As a result, constructing a straight line between the initial and final poses of the exoskeleton does not always result in an MDC path for the human arm [11]. This issue prevents the direct use of computational models formulated in the c-space of the human arm for motion generation in exoskeletons.

This paper introduces a framework for solving the problem of human-like motion generation in upper-limb exoskeletons. The proposed framework is a two-stage process where a human-like reference path is generated in the configuration space of the human arm first. Supporting Scapulohumeral Rhythm (SR) and spatial similarity of the generated motions to the healthy motion patterns (in task space as well as human joint space) are the human-like qualities considered. To this end, first a computational model is developed by integrating inner shoulder models into the minimum kinetic energy path-generation algorithm in Riemannian space. The second stage involves transforming the generated path using an exoskeleton-specific transformation developed for mapping the c-space of the human arm to the exoskeleton’s c-space. This mapping between the two spaces is derived using analytical geometry and is named Geometric Equivalence for Anthropomorphic Arms (GEAA). The proposed path generation approach is applied to a six degrees-of-freedom upper-limb exoskeleton (CLEVERarm [29]) which supports scapulohumeral rhythm, humeral rotations and elbow flexion/extension. To validate the effectiveness of the proposed GEAA transformation, it is shown that kinematic characterizations of the transformed trajectory and the outputs of the underlying computational model are equivalent. Additionally, through an experimental approach, it is shown that the outputs of the proposed framework bear a strong resemblance to the natural motions of the arm. It is important to note that the proposed method can be adapted for use on other upper-limb exoskeleton with different kinematic structure and supports the use of different computational models. This paper is organized as follows: Section 2 presents the terminology used and outlines the two-stage path generation algorithm in a conceptual level, Section 3 presents the motion planning in human configuration space, Section 4 demonstrates construction of the GEAA mapping for a six degree of freedom exoskeleton. The data collection procedure, simulation and experimental results, and quantitative analysis of the acquired data are presented in Section 5. Finally, Section 6 concludes the paper by reviewing the main contributions and findings of the paper.