Integration of neuromuscular control for multidirectional horizontal planar reaching movements in a portable upper limb exoskeleton for enhanced stroke rehabilitation

Introduction

With the global demographic trend towards an aging population, healthcare systems across the world face increasing pressures to address the needs of elderly individuals [1]. Stroke, a primary cause of mortality and disability among older adults [2], frequently leads to upper limb hemiparesis due to damage in the central nervous system [3]. This condition not only impairs motor functions but also imposes substantial financial and caregiving burdens on families and communities. Although traditional manual rehabilitation has proven effective for hemiplegic patients [4], its high costs and limited adaptability to diverse patient requirements restrict its broader application [5], [6], [7].

To address the disparity between the high prevalence of stroke and the scarcity of professional rehabilitation therapists, recent research has explored wearable exoskeleton robots as a feasible alternative [8]. These patients often require extended rehabilitation periods; hence, portable exoskeletons enable therapy in home settings or during daily activities, reducing the need for frequent clinical visits [9]. This method enhances treatment frequency and duration, thereby expediting recovery. Additionally, the exoskeletons’ lightweight construction lessens patient strain, increases comfort, and promotes sustained usage [10]. Consequently, research is shifting from traditional, stationary upper-limb rehabilitation exoskeletons towards the development of portable models adaptable to various environments [11].

Research on portable upper limb rehabilitation exoskeletons is divided into five primary categories: electrically driven, pneumatically driven, cable-driven, soft, and hybrid exoskeletons [11], [12], [13], [14], [15]. Cable-driven and soft exoskeletons, known for their lightweight and comfortable design, facilitate easy donning and doffing, making them suitable for extended use [16]. However, their limited power output constrains their application in rehabilitation contexts [17]. In contrast, pneumatically driven exoskeletons, which offer high compliance and adaptability, are well-suited for complex movement assistance in rehabilitation [18]. Nevertheless, their dependence on external air sources and noise production significantly diminishes their portability and increases environmental dependency [19]. Hybrid exoskeletons, which combine multiple driving mechanisms, provide tailored adaptability for specific rehabilitation needs [20], albeit at the cost of increased system complexity, higher expenses, and the need for precise control systems [21]. Considering the balance between portability, cost, and rehabilitation needs, electrically driven exoskeletons stand out as the optimal choice. They offer significant power and precise control, ideal for assisting rehabilitation training, and feature an integrated battery design that enhances portability. Moreover, selecting appropriate motors can effectively manage costs, aligning with economic considerations [22].

Therefore, this study introduces an electrically driven, portable upper limb exoskeleton system specifically designed to address the highlighted challenges for stroke patients. This system integrates the neuromuscular control mechanisms of planar reaching movements of the human arm, aiming to facilitate rehabilitation exercises in a natural and effective manner. The primary contributions of this paper are summarized as follows:

1. Innovative Integration of Neuromuscular Control Mechanisms: For the first time, the control system of an upper limb rehabilitation exoskeleton incorporates the neuromuscular control mechanism associated with Multidirectional Horizontal Planar Reaching Movements of the human arm. This integration enables the exoskeleton to drive the arm in naturalistic movements, thereby activating cerebral and neural pathways, promoting neurological reorganization. Such movements are crucial for maintaining and enhancing muscle strength, coordination, and endurance, and are essential for functional recovery.

2. Structural Design and Material Utilization: To balance lightness with structural integrity, the exoskeleton utilizes PLA material fabricated using 3D printing technology. A dual-layer design at critical points further enhances structural strength, ensuring the exoskeleton remains portable while not significantly compromising the system’s structural integrity.

3. User-Friendly Human-Machine Interface: The exoskeleton is equipped with a sophisticated human-machine interface that displays real-time data of key performance metrics, including joint angles, rotational speeds, and torques of the upper limb. This interface enhances user interaction, allowing for immediate feedback and adjustment, which is critical for effective rehabilitation.

4. Cost-Effectiveness Analysis: The designed upper limb exoskeleton system ensures that the motors provide stable torque, facilitating the simulation of natural planar extension movements of the human arm and promoting neuroplasticity and functional recovery. The overall cost of the system has been effectively managed, making it more cost-effective compared to other upper limb exoskeleton systems on the market.

The structure of the remainder of this paper is as follows: Section “Neuromechanical modeling” provides a detailed description of the kinematic and dynamic modeling of human skeletal muscle arms during planar motion, along with the modeling of neural control mechanisms for these movements. Section “Exoskeleton design” introduces the design of the portable upper-limb rehabilitation exoskeleton developed in this study, covering aspects such as the mechanical structure design, motor and drive systems, embedded control systems, and the human-machine interface. Section “Results” presents the experimental results, including both simulations and real-world testing of the exoskeleton. Section “Discussion” discusses the advantages of this research, such as cost control, addresses its limitations, such as the customization of the product which may not be suitable for all stroke patients, and explores the impact of this work on the field. Finally, Section “Conclusions” concludes the paper.

Neuromechanical modeling

Neuromechanics integrates knowledge from neuroscience, biomechanics, and physiology to study how the nervous system controls bodily movements [23]. Neuromechanical modeling, a significant branch of neuromechanics, employs mathematical and computational models to simulate and analyze the interactions between the nervous and muscular systems that result in human motion [24]. By integrating these neuromechanical models into the design of exoskeleton controllers, it becomes possible to provide consistent and repetitive training in a patient’s initial natural motion patterns [25]. This approach not only simulates these movements but also leverages neural plasticity to reshape the brain regions responsible for motor control, thereby accelerating the rehabilitation process [26].

Musculoskeletal model

A key component of one neuromechanical model is the accurate representation of the biomechanical systems involved. For instance, in studies of multidirectional planar reaching movements of the human arm, a detailed musculoskeletal model of the human arm was developed, as depicted in Figure 1. The construction and validation of this model have been substantiated in our previous research [23], 27], 28].

Figure 1:

Constructed human musculoskeletal arm model: (a) Planar skeletal arm model consisting of two skeletal links; (b) planar musculoskeletal arm model featuring six labeled muscles: l ₁ (SF), l ₂ (SX), l ₃ (EF), l ₄ (EX), l ₅ (BF), and l ₆ (BX).

Kinematic model

The kinematics of the human arm are analyzed within the Cartesian coordinate system. Based on this analysis, the end position of the arm, hand, in Cartesian space can be expressed as follows:

(1) p = x y = L 1 cos θ 1 + L 2 cos θ 1 + θ 2 L 1 sin θ 1 + L 2 sin θ 1 + θ 2

where the vector p represents the position of hand and L ₁ and L ₂ denote the lengths of the upper arm and the forearm, respectively. The angle θ ₁ is defined as the angle between the upper arm and the X-axis, and θ ₂ is the angle between the upper arm and the forearm. The joint angles, θ ₁ and θ ₂, can be inferred from the hand position.

(2) θ = θ 1 θ 2 = arctan y x − arctan L 2 sin θ 2 L 1 + L 2 cos θ 2 arccos x 2 + y 2 − L 1 2 − L 2 2 2 L 1 L 2

By differentiating Equation (1), the expressions for linear velocity p ˙ and angular velocity θ ˙ can be described as follows:

(3) p ˙ = J θ

(4) p ˙ = x ˙ y ˙

(5) J = − L 1 sin θ 1 − L 2 sin θ 1 + θ 2 − L 2 sin θ 1 + θ 2 L 1 cos θ 1 + L 2 cos θ 1 + θ 2 L 2 cos θ 1 + θ 2

(6) θ = θ 1 θ 2

where J denotes Jacobian matrix, performing the transformation of joint velocities into hand velocities.

Subsequently, the effects of muscle contraction were incorporated into the kinematic model. It was assumed that the muscle contractions were linear and the impact of mass transfer during contraction was negligible. Consequently, the muscle lengths could be defined by the vector l = l 1 , l 2 , l 3 , l 4 , l 5 , l 6 T . The representations of these lengths are detailed below.

(7) l 1 l 2 l 3 l 4 l 5 l 6 = a 1 2 + b 1 2 + 2 a 1 b 1 cos θ 1 1 2 a 2 2 + b 2 2 − 2 a 2 b 2 cos θ 1 1 2 a 3 2 + b 3 2 − 2 a 3 b 3 cos θ 2 1 2 a 4 2 + b 4 2 − 2 a 4 b 4 cos θ 2 1 2 c 51 2 + c 52 2 + L 1 2 + 2 c 51 L 1 cos θ 1 + 2 c 52 L 1 cos θ 2 + 2 c 51 c 52 cos θ 1 + θ 2 1 2 c 61 2 + c 62 2 + L 1 2 − 2 c 61 L 1 cos θ 1 − 2 c 62 L 1 cos θ 2 + 2 c 61 c 62 cos θ 1 + θ 2 1 2

where the coefficients a _i (i=1, 2, 3, 4), b _j (j=1, 2, 3, 4), a ₅₁, a ₅₂, a ₆₁, a ₆₂, b ₆₁, and b ₆₂ represent the moment levers in the mechanical model. These coefficients are assumed to be constants, invariant throughout the arm’s movement. The matrix representation of these moment levers is provided below.

(8) q = − a 1 a 2 0 0 − c 51 c 61 0 0 a 3 a 4 − c 52 c 62 T

Similarly, the linear velocity of l can be obtained by differentiating Equation (7).

(9) l ˙ = W θ ˙

where W denotes the Jacobian matrix.

Dynamic model

Considering that the reaching movements of the human arm occur predominantly in a plane, the effects of gravity are omitted from this analysis for simplification. Consequently, using the Lagrangian equation allows for the formulation of the dynamic equations of the skeletal system, as presented below.

(10) M ( θ ) θ ˙ + C ( θ , θ ˙ ) θ ˙ + V θ ˙ = τ

where the matrices M , C , and V represent the inertia, Coriolis, and viscosity matrices, respectively. Each matrix is detailed as follows:

(11) M ( θ ) = α + 2 β cos θ 2 δ + β cos θ 2 δ + β cos θ 2 δ

(12) C ( θ , θ ˙ ) = − β θ ˙ 2 sin θ 2 − β θ ˙ 1 + θ ˙ 2 sin θ 2 β θ ˙ 1 sin θ 2 0

(13) V = v 11 v 12 v 21 v 22

where the elements of these matrices are defined as follows:

(14) α = I 1 + I 2 + m 1 L g 1 2 + m 2 L 1 2 + L g 2 2

(15) β = m 2 L 1 L g 2

(16) δ = I 2 + m 2 L g 2 2

where m ₁ and m ₂ denote the masses of the upper arm and forearm, respectively. I ₁ and I ₂ represent the moments of inertia for these segments. L ₁ and L ₂ refer to the lengths of the upper arm and forearm, respectively, while L _g1 and L _g2 denote the distances from the joints to the centers of mass of each segment.

Given the pronounced nonlinear behavior of skeletal muscles, the Hill-based muscle model has been adapted to better capture the force dynamics of these muscles, as shown in Figure 2. This model primarily consists of the contractile component (CC), representing the active force production in response to muscle excitation, and the series elastic component (SEC), which models the elastic force response. In this revision, the damping effect has been emphasized by setting the elastic coefficient of the SEC to a near-infinite value, thereby enhancing the model’s ability to simulate scenarios where damping effects are predominant [29].

Figure 2:

Hill-based muscle model schematic: diagram of the muscle model, including the CC, PEC, and SEC.

According to the classical Hill muscle model, which primarily describes the force-velocity relationship of muscle fibers, the neural control signal is often represented as a normalized activation level within a range of 0–1. This normalized signal, a summation of the motor unit action potentials, serves as the input for muscle contraction. Upon reception by the CC of sarcomeres, this signal undergoes a transformation to an activation state that also ranges from 0 to 1. The conversion time is linked to the dynamics of chemical reactions involved in muscle contraction, specifically the release and reuptake of calcium ions that facilitate the cross-bridge cycling between actin and myosin filaments. To quantitatively describe this process, the contractile force generated by a muscle can be mathematically expressed as follows:

(17) f c c = f 0 ⋅ α ⋅ f fl ⋅ f f v

where f ₀ represents the maximum isometric force, while f _fl and f _fv denote the nondimensional factors corresponding to the force-length and force-velocity relationships, respectively. Furthermore, the variable α indicates the activation level.

(18) f s e c = γ l s e c − l u    if  l s e c > l u 0    if  l s e c ≤ l u

(19) γ = f 0 u 0 l 0 2

where l _sec denotes the length of the SEC, while l _u represents its unloaded length. The constant u ₀, generally assigned a value of 0.04.

(20) f p e c cos α = f s e c − f c c cos α

(21) f t = f p e c + f s e c = f p e c + f c c cos α

where f _t denotes the total force exerted on the muscle tendon.

Neural control model

To align with the natural kinematics of human motion, an optimal control strategy was developed that accounts for both the biomechanical constraints of the human body and the physiological characteristics of muscle activation. The process of neuromuscular control is depicted in Figure 3. By incorporating position, velocity, force, and muscle activation parameters into the cost function, the effectiveness and comfort of the rehabilitation process can be significantly improved. The formulation of the cost function is presented as follows:

(22) c o s t = w p p ( T ) − p * 2 + w v ‖ v ( T ) ‖ 2 + w f ‖ f ( T ) ‖ 2 + ∫ 0 T ‖ u ( t ) ‖ 2 d t

where T represents the terminal time, p * denotes the target position, and p (T), v (T), and f (T) represent the terminal position, velocity, and force at the endpoint of the human arm, respectively. The weights w _p, w _v, and w _f are assigned to each term accordingly, while u (t) indicates the motor unit’s action.

Figure 3:

Neuromuscular control process block diagram: depicts the sequence of neuromuscular control, beginning with brain signals that initiate muscle contractions, ultimately resulting in movement generation.

Through the integration of the iterative linear-quadratic-Gaussian approach with the Levenberg-Marquardt algorithm, the cost function was refined to ascertain the optimal u (t), applying it to generate natural and smooth movement trajectories [30]. These trajectories are reevaluated once the initial and target positions are established. Subsequently, the exoskeleton adheres to these trajectories to ensure that the rehabilitation process is aptly suited for stroke patients.

Exoskeleton design

The portable upper limb exoskeleton developed in this study is composed of five primary components: (1) two arm brackets that support and immobilize the upper limb and forearm; (2) a rear frame support that houses the embedded system, battery pack, and body attachment mechanisms; (3) three DC motors that facilitate the control of the elbow and shoulder joints; (4) an embedded system programmed to process joint angle data and dispatch control signals to the motors; and (5) a battery pack that supplies power to both the motors and the embedded system. A detailed illustration of the exoskeleton design is presented in Figure 4.

Figure 4:

Exoskeleton design visualization: (a) 3D schematic of the exoskeleton, showing its overall structure; (b) exploded view detailing individual components and their assembly; (c) photograph of the actual exoskeleton, demonstrating its real-world appearance and scale.

Mechanical structure design

Figures 5 and 6 depict the mechanical design of the forearm and upper limb brace. The exoskeleton, fabricated using 3D-printed polylactic acid (PLA), exhibits non-toxicity, superior mechanical properties, low density, and excellent bioaffinity. To minimize weight while preserving structural integrity, the bracket’s surface incorporates three strategically placed holes. These apertures not only reduce the overall mass, thereby enhancing the system’s dynamic properties, but also function as versatile mounting points for sensors. This design feature facilitates easy sensor interchangeability, accommodating various research emphases. Owing to its reduced mass and optimized kinematics, this exoskeleton design requires less motor torque compared to traditional metal counterparts, thus allowing the use of smaller, lighter motors. Additionally, the employment of PLA contributes to environmental sustainability, as the material is both recyclable and biodegradable. The integration of straps and the bracket’s design effectively prevents the user’s arm from slipping.

Figure 5:

Upper limb bracket representation: 3D model of the upper limb bracket at the center, with top, front, and side view drawings surrounding it.

Figure 6:

Forearm bracket representation: 3D model of the forearm bracket at the center, accompanied by top, front, and side view drawings.

Figure 7 illustrates the back support system of the exoskeleton. Constructed from aluminum alloy, the rear brace is notably lightweight, enhancing user comfort. To optimize balance, the battery pack and controller are strategically positioned at the back, effectively counterbalancing the frontal arm brace and motor. This arrangement reduces balance strain during movement. The rear bracket is also engineered to include a dedicated compartment for the integrated controller and battery pack, facilitating maintenance and upgrades. Its modular design supports size customization without incurring additional costs. Furthermore, the rear brace is secured to the user with two adjustable straps, and soft fabric padding is incorporated to augment comfort and support.

Figure 7:

Back brace representation: 3D model of the back brace in the center, with corresponding top, front, and side view drawings.

Motor and drive system

Figure 8 elaborates on the motor design used in the exoskeleton [31]. This motor provides increased torque relative to previous exoskeleton motors while maintaining a compact form factor, thereby enhancing portability. The gear reduction ratio is set at 1:6, which not only mitigates the risk of injury due to the motor’s self-locking capability but also allows users to exert rotational force even in the absence of power. Furthermore, the design accommodates the application of greater reverse torque by the user during operation, supporting resistance in rehabilitation movements as well as assisted motion.

Figure 8:

DC motor depiction: (a) exploded view of the DC motor, detailing internal components and assembly structure; (b) and (c) three-view drawings of the motor from the left and right sides, respectively.

Embedded control system

The STM32F407ZGT6 microcontroller functions as the central processing unit (CPU) within our exoskeleton system, executing the compiled code and overseeing system operations. It features a high-performance Arm Cortex-M4 32-bit RISC core, capable of operating at frequencies up to 168 MHz. The Cortex-M4 core includes a single-precision floating-point unit (FPU), supporting all ARM single-precision data-processing instructions and data types. Additionally, this microcontroller offers high-speed embedded storage, comprising up to 1 M bytes of flash memory and 192 Kbytes of SRAM, along with enhanced I/O capabilities and extensive peripheral support. In our application, the microcontroller facilitates the analysis of forward and inverse kinematics and dynamics at computation rates up to 1,000 Hz, significantly enhancing operational efficiency.

A custom printed circuit board (PCB) was designed to integrate sensors, actuators, and the microcontroller effectively. This 120 × 89 mm circuit board utilizes Japan Solderless Terminal (JST) connectors for connecting all sensors and actuators. Beyond the standard STM32F407 features, our PCB includes a power supply module, a Joint Test Action Group (JTAG) debug interface, various communication modules, and operational keys. Figure 9 illustrates the primary components of the PCB. The power supply module adjusts the external power source to the necessary voltage levels for each component, accommodating the exoskeleton’s 24 V operational voltage. For online debugging and programming, a JTAG debug interface is implemented. The communication capabilities are extensive, featuring RS-232, RS-485, IO, CAN-BUS, USART, high-speed USB, and RS-485. Three motor drivers are linked to the CAN-BUS module, managing motor torque, rotation angle, and speed. Additionally, the wireless module, connected to the USART1, transmits data at a 460,800 baud rate to ensure signal stability. Most keys provide backup functionality, with one dedicated to initiating the RESET signal.

Figure 9:

Real-life PCB image: photograph showing the actual PCB used in the exoskeleton, highlighting key components.

Human-machine interface

A HMI was designed to provide users with an intuitive means of observing and controlling their movement status, facilitating data visualization. As depicted in Figure 10a, various data types, such as velocity, angular velocity, and joint torque, are displayed by the HMI. Notably, it is not limited to presenting the motor’s parameters; physical metrics of the user’s arm, as measured by sensors, are also shown. This feature allows for the comparison between the arm’s angle and the motor’s angle, aiding users in assessing the appropriateness of their movements. Real-time commands can be sent via the HMI to control the exoskeleton, assisting users in their rehabilitation. Additionally, personalization features are offered by the HMI, enabling users to tailor the interface to highlight specific data or present it in their preferred format. As illustrated in Figure 10b, data is represented in diverse forms such as curves, bars, and instrument pointers, with the theme of the HMI set to night mode.

Figure 10:

HMI display of movement information in different modes: (a) HMI in light mode, showing real-time movement data including position, velocity, and torque, for the current rehabilitation exercise. After completing the 32-target position-reaching task, all information can be compiled into the results shown in Figures 12 and 13; (b) same data visualization in dark mode, demonstrating the interface’s readability in various lighting conditions.

Results

To enhance the efficacy of rehabilitation training for stroke patients, the control system of our newly designed portable upper limb exoskeleton has been integrated with the natural neural control mechanisms governing planar human arm reaching movements, as depicted in Figure 11. This integration aims to emulate the movement characteristics of healthy individuals by simulating normal motor patterns, thereby facilitating neural recovery and enhancing muscular control. Such an approach is corroborated by current rehabilitation theories which advocate for mimicking healthy human motion patterns as this method has been shown to support more natural and effective restoration of motor functions in patients.

Figure 11:

Integration of neuromuscular controller into exoskeleton control framework: illustration of how natural movement trajectories generated by the neuromuscular controller for the musculoskeletal arm model are transmitted to the exoskeleton’s embedded system. The system computes inverse kinematics and determines the torque required by the motors to drive the exoskeleton along predetermined trajectories. Motor encoders provide real-time feedback on angles and angular velocities, which are adjusted by a classical PD controller to ensure accurate movement, facilitating effective rehabilitation.

The simulation results of the musculoskeletal arm model, optimized by a neural controller based on a predefined cost function, are illustrated in Figure 12. Over the simulation timeframe, it was demonstrated that the arm’s endpoint accurately reached the target positions in 32 distinct directions, originating from a central point and radiating outward in equally spaced directions. The trajectories displayed smooth, natural curves, and the velocity profiles of these movements typically followed bell-shaped curves – starting from zero, peaking, and then returning to zero. Additionally, the joint torques exhibited patterns of positive and negative peaks, consistent with the typical planar movements of a healthy human arm, thereby indicating the successful replication of natural movement patterns.

Figure 12:

Simulation results of the musculoskeletal arm model under neuromuscular control: visualization of the simulation outcomes for the arm model, controlled by a neural controller optimized based on a predefined cost function. (a) Trajectories of 32 paths from the center to target points, evenly distributed at 360° intervals; (b) corresponding velocity profiles for each of the 32 trajectories; (c) force magnitude of hand for each of the 32 trajectories.

These simulation insights allow for the customization of the musculoskeletal arm model parameters to suit the unique arm characteristics of individual stroke patients. By designing tailored movement trajectories, the exoskeleton can guide patients through personalized rehabilitation sessions. Figure 13 displays the exoskeleton’s performance, demonstrating its ability to accurately follow the simulated trajectories across 32 distinct paths. The results confirm the exoskeleton’s high precision in replicating the planned movement paths. Figure 14 shows the exoskeleton’s accuracy in following simulated trajectories across 32 distinct paths. The box plots illustrate both the average and maximum errors in the X-axis, Y-axis, and overall directions. The results confirm the exoskeleton’s high precision in replicating the planned movement paths. Specifically, the average error across all trajectories is approximately 1 mm, while the maximum error is around 2 mm, demonstrating the system’s ability to accurately follow the desired movement trajectories and ensure effective rehabilitation exercises.

Figure 13:

Detailed comparison of exoskeleton performance on selected trajectories: comparative analysis of four selected trajectories, showing the exoskeleton’s movement paths alongside simulated trajectories.

Figure 14:

Box plots illustrating the accuracy of the exoskeleton in replicating the simulated musculoskeletal arm model’s movement trajectories. (a) Average error values for the X-axis, Y-axis, and overall directions across 32 simulated movement trajectories. (b) Maximum error values for the same categories.

Discussion

Compared to traditional pedestal-mounted upper limb exoskeleton systems commonly found in professional medical institutions, such as ETS-MARSE [32], ground-based upper limb exoskeletons are generally bulkier and more complex. These systems incorporate additional support structures and joint drivers to deliver enhanced force and stability. Due to the requirement for more materials and intricate mechanical designs, these systems are associated with higher costs. Furthermore, ground-based systems can provide greater force output and more precise control, which are critical for certain rehabilitation therapies or heavy-duty tasks. To achieve these performance metrics, the systems are equipped with higher specification motors and control systems, components that are costly, driving the overall hardware cost to approximately $18,000 USD. For instance, the Co-Exos II exoskeleton builds upon this by upgrading the lifting base [33], resulting in a more complex structure and an increase in machined parts, which raises the estimated hardware cost to about $23,000 USD. In contrast, portable exoskeletons like the CURER [34], a Cable-Driven Compliant Upper Limb Rehabilitation Exoskeleton Robot, require high-precision synchronization of cable drums and connection components to ensure smooth operation and patient safety. Achieving this precision typically necessitates customized design and precision machining, significantly increasing costs, with an estimated hardware expense of around $7,000 USD. These costs are much higher than those of the exoskeleton proposed in this study, as shown in Table 1.

Although the upper limb exoskeleton designed in this study shows satisfactory performance in terms of portability, cost control, and human-machine interaction, its comfort in wearability remains suboptimal. To balance cost and portability while supporting the entire weight of the exoskeleton, aluminum alloy was chosen as the primary material. However, the decision to use prefabricated components from factories to further control production costs has limited personalized design adjustments for the back contour, adversely affecting comfort levels.

Additionally, as the exoskeleton was customized based on the physical parameters of the author, its generalizability is limited. Future research should focus on enhancing the adjustability of the exoskeleton to better accommodate a wider range of users. Furthermore, this study primarily utilizes the neuromuscular control mechanisms of multidirectional horizontal planar reaching movements to facilitate arm movements in stroke patients. However, it is crucial to recognize that in more complex upper limb rehabilitation exercises, hand movements cannot be overlooked. In scenarios involving intricate arm movements, akin to spatial manipulation tasks, failing to consider the intricacies of hand function may result in non-learning use of the hand, leading to actions that are functionally ineffective and practically useless.

Overall, one of the primary innovations of this paper is the integration of neuromuscular control strategies into exoskeleton control. The combination of neuromuscular control with exoskeletons is not only applicable to stroke rehabilitation training but also extends to a broader range of applications. For instance, this technology can calculate the forces exerted by the exoskeleton based on feedback signals, thereby assessing the risk of muscle strain for the wearer and preventing medical incidents. Additionally, this control strategy is suitable for enhancing athletic performance. In the absence of real-time feedback, it can predict future actions based on previous movements, thus enhancing human-machine interaction and improving movement efficiency. Therefore, the integration of neuromuscular control with exoskeletons is expected to have extensive implications for future applications.

Conclusions

This paper introduces a novel upper limb exoskeleton designed for stroke rehabilitation, aimed at assisting physical therapists in restoring motor functions of patients. Addressing the needs of early-stage stroke patients, the exoskeleton was constructed using 3D printing technology with PLA material, ensuring it is lightweight and portable. A dual-layer design was implemented to enhance structural integrity and load-bearing capacity, optimizing the balance between weight and strength. This design allows patients to engage in rehabilitation in various daily scenarios, facilitating sustained and prolonged training sessions. The exoskeleton incorporates an embedded system that controls electric motors to deliver consistent torque, modulated by a neural controller developed through optimization algorithms. This ensures a patient-centered approach to rehabilitation. Simulation results indicate that the neural controller effectively generates bell-shaped velocity curves and smooth endpoint trajectories with coordinated antagonist muscle activity, mirroring the critical features of natural planar extension movements. Experiments with the real exoskeleton-driven arm have confirmed these characteristics, further validating the effectiveness of the design and methodology.