A multi-DOF robotic exoskeleton interface for hand motion assistance

This paper outlines the design and development of a robotic exoskeleton based rehabilitation system. A portable direct-driven optimized hand exoskeleton system has been proposed. The optimization procedure primarily based on matching the exoskeleton and finger workspaces guided the system design. The selection of actuators for the proposed system has emerged as a result of experiments with users of different hand sizes. Using commercial sensors, various hand parameters, e.g. maximum and average force levels have been measured. The results of these experiments have been mapped directly to the mechanical design of the system. An under-actuated optimum mechanism has been analysed followed by the design and realization of the first prototype. The system provides both position and force feedback sensory information which can improve the outcomes of a professional rehabilitation exercise.


I. INTRODUCTION
ANDS are central entity for the maintenance of independent living. Hand exoskeleton systems enhance the human strength by extending the sensing and manipulation capabilities of a user in a real and/or virtual environment. In rehabilitation, the primary aim of such systems is to assist physiotherapists in performing the therapies after hand injuries or strokes, thereby partially or even completely replacing the classical manual procedures. Construction of assistive systems for the benefit of humankind has always fascinated the research community. The scientific literature reports many exoskeleton-based hand rehabilitation systems.
Scientists at HongKong Polytechnic University have conceived a complete five-fingered hand with 2 DOF per finger employing Virtual Centre of Rotation (VCR) mechanism [1]. An exoskeleton system having 1 DOF has been presented by KAIST researchers considering Activities of Daily Living (ADL) training for stroke patients [2]. The main objective was to realize several types of grasp including cylindrical, lateral and pinch. A 4-bar linkage has been designed to imitate the finger tip path in grasping motion while a cable mechanism drives the movement of the thumb. Another exoskeleton aimed to restore dexterity of paralyzed hands has been developed at CMU [3] [4]. The system is actuated by four actuators placed at a distance from the hand to reduce loading. The exoskeleton finger is comprised of three phalanges corresponding to the human hand anatomy. For each finger joint, two cables have been used to transmit force and motion from the actuator to the exoskeleton. Another novel hand exoskeleton exerciser comprised of four fingers having 7 active DOF has been proposed by researchers at Salford University [5]. The system has been intended to combine dexterity with a good Range Of Motion (ROM). The actuators reside on ground and the bidirectional forces are transmitted by low friction tendons. The device has been integrated within a Virtual Reality (VR) based hand therapy system, thus permitting a clinician to customise and perform hand exercises and finger motion evaluation tests. Another tendon-driven hand exoskeleton having 4 DOF per finger has been conceived by Wege et al. [6]. The system is capable of exerting bidirectional forces using a single DC motor. Researchers at SSSA, Italy have realized a novel hand exoskeleton system [7] intended to simplify the exoskeleton complexity related to its structure, mechanism and actuation while still providing full hand mobility. The natural ROM has been accomplished by keeping the number of exoskeleton's DOF similar to that of a natural hand while simplicity has been achieved by proposing a novel mechanical design. Y. Fu  An earlier developed Hand EXOskeleon SYStem (HEXOSYS)-I can provide force levels (45N) beyond any existing system [9]. The experiences with HEXOSYS-I enabled us to design a new light mass, less volumetric hand exoskeleton that is more ergonomic thus yielding better performance in terms of grasping and manipulation. Moreover, the newly developed exoskeleton can accommodate up to 5 fingers and supports adjustment of various hand sizes. While both the proposed hand exoskeleton systems are direct-driven and portable with the ability to exert bi-directional forces on the finger phalanges, their mechanisms, actuation systems, optimization criteria and physical features are entirely different.
The design concept of HEXOSYS-II finger is presented in Figure 1. With an underactuated mechanism, it is a two link serial Revolute Revolute (RR) manipulator which is attached to the finger at a single point. The system is powered by a single actuator residing at the base of exoskeleton's proximal joint. The finger prototype is shown in Figure 2. The functional behavior of an exoskeleton in the WorkSpace (WS) strongly depends on the lengths and shape of its links. This motivated us to carry a multi-parametric optimization procedure that determines the optimized link lengths. The optimization criteria primarily include fingerexoskeleton WS matching in addition to other trivial factors like kinematic mapping, worst case collision avoidance, etc. For the sake of widening the reachable exoskeleton WS without encountering the collision, the first link of length L1 (see Figure 1) has been split into sub-segments (Figure 3). The lengths of these segments are adjustable and are a function of finger and hand size. Likewise, the angle between the first two sub-segments (θ fixed ) is also adjustable but is fixed for a certain hand/finger. The length of the distal link (L2) which serves as a connection link between the finger and the end-effector has been set to the minimum possible allowed by the mechanical integration (1cm). The optimization algorithm starts with assuming reasonable lengths of the segments. Each set of link lengths is then subjected to traverse throughout the finger WS for analysis. The finger WS has been determined using the Monte Carlo method. Random samples of the finger joint angles in the range -10°≤θ MCP ≥50°, 0°≤ θ PIP ≥110° determine the points throughout the finger WS using Where L 1f and L 2f are the lengths of proximal and middle digits of the human finger respectively while C and S refer to Cosine and Sine of the corresponding angles respectively.
The set of segment lengths is analyzed to determine the number of points in the finger WS reachable by the exoskeleton without collision. For collision detection, the An exoskeleton link length set is considered as collision-free if all the points on the links reside outside the rectangular envelopes. The collision-free WS is stored for comparison with the next iterated link lengths set. When all the segment lengths are iterated, the set giving maximum correlation of the finger WS and the exoskeleton WS is considered as optimized. Figure 4 illustrates flow-chart of optimization procedure.
In case of an index finger of a medium sized hand, the optimized segment lengths as found from the algorithm are L1-1=8cm, L1-2=2cm, L1-3=2cm with θ fixed =55.4°. The finger WS and the exoskeleton WS corresponding to these optimized link lengths are illustrated in Figure 5. This shows that the optimized RR mechanism fully covers the natural ROM of a finger. The HEXOSYS-I was targeted at maximum force capabilities of the human hand. In an attempt to reduce the system physical dimensions and thus enhancing the ergonomics, HEXOSYS-II has been aimed at exerting average force levels. These levels measured with various devices including force sensors and the load cell can be ultimately mapped to lower level requirements such as actuation torque. Three healthy subjects each having small, medium and big hands participated in the experiments. One of the experiments included recording the force levels required to accomplish some usual grasping activities. The subjects were asked to grasp and manipulate the objects in the same fashion as they interact with them in their daily lives. The commercial FingerTPS™ Tactile Pressure Sensors have been used to measure the force exerted by the finger tips. Fig. 6 and 7 show the force profiles of two activities in case of Right (Rt.) thumb, index and middle digits. Detailed design requirements have been reported in [10]. The optimized link lengths and shape presented in Section III guided the design of the HEXOSYS-II structural subsystem while the results of average force measurement experiments (Section IV) paved the way to choose actuators for the proposed system.
The exoskeleton finger consists of an actuator per finger together with its accessories, a pair of bevel gears, optimized links and sensors. An exploded CAD view of a single exoskeleton finger is illustrated in Figure 8. The actuator is a DC motor by Portescap (16G88-220P). It can provide torque up to 16mNm and has a mass of 37gm including the gear, thus making the system light weight. The actuator accessories consist of a 2 stage planetary gear-head with ratio of 30.2:1 and a Magneto-Resistive (MR) encoder with a resolution of 512 pulses per revolution. The use of bevel gears, by changing the orientation of motor axis permits the extension of exoskeleton fingers. A miter gear pair (1:1) from Boston Gear made up of stainless steel has been used in every exoskeleton finger. Two types of sensors mounted on the dev the device motions and interaction forces. force sensor has been mounted imme abduction-flexion connector (shown in yell The sensor based on strain gauge measure forces with the finger segment. In additi encoder, a 12-bit programmable magnetic available from Austriamicrosystems has b the passive revolute joint to measure the p link joints. This permits monitoring of the which the fingertip posture can also be means that no additional instrumentation is the finger motions, e.g. data-gloves. The C complete HEXOSYS-II prototype is illustra The links and the base of the exoske fabricated using a high-tech in-house 3D p plastic to reduce the overall weight of custom miniature parts (e.g. pins, lock fabricated in steel while the rest of the cu been made up of light aluminium. A par HEXOSYS-II finger prototype is presente The base of index, middle and ring fingers ded view vice measure both A custom-made ediately on the low in Figure 8). es the interaction ion to the motor c rotary encoder been mounted on position of the 2exoskeleton from e extracted. This required to track CAD model of the ated in Figure 9. el eleton have been printer using ABS the system. The kers) have been ustom parts have rtially assembled ed in Figure 10. s is also shown in the figure. The proposed concept i each finger individually and pr extension as well as passive abductio Fig. 10. HEXOSYS-II p