A Synergistic Rehabilitation Approach for Post-Stroke Patients with a Hand Exoskeleton: A Feasibility Study with Healthy Subjects

WOW! You got published by faking stroke research on healthy subjects? You should all be fired for cause!

A Synergistic Rehabilitation Approach for Post-Stroke Patients with a Hand Exoskeleton: A Feasibility Study with Healthy Subjects

,

,

,

and

1

Institute of Mechanical Intelligence, Scuola Superiore Sant’Anna, 56127 Pisa, Italy

2

Department of Engineering and Science, Universitas Mercatorum, 00186 Rome, Italy

*

Author to whom correspondence should be addressed.

Robotics2026, 15(1), 21;https://doi.org/10.3390/robotics15010021

This article belongs to the Special Issue AI for Robotic Exoskeletons and Prostheses

Version Notes

Order Reprints

Abstract

Hand exoskeletons are increasingly used to support post-stroke reach-to-grasp, yet most intention-detection strategies trigger assistance from local hand events without considering the synergy between proximal arm transport and distal hand shaping. We evaluated whether proximal arm kinematics, alone or fused with EMG, can predict flexor and extensor digitorum activity for synergy-aligned hand assistance. We trained nine models per participant: linear regression (LINEAR), feedforward neural network (NONLINEAR), and LSTM, each under EMG-only, kinematics-only (KIN), and EMG+KIN inputs. Performance was assessed by RMSE on test trials and by a synergy-retention analysis, comparing synergy weights from original EMG versus a hybrid EMG in which extensor and flexor digitorum measure signals were replaced by model predictions. Results have shown that kinematic information can predict muscle activity even with a simple linear model (average RMSE around 30% of signal amplitude peak during go-to-grasp contractions), and synergy analysis indicated high cosine similarity between original and hybrid synergy weights (on average 0.87 for the LINEAR model). Furthermore, the LINEAR model with kinematics input has been tested in a real-time go-to-grasp motion, developing a high-level control strategy for a hand exoskeleton, to better simulate post-stroke rehabilitation scenarios. These results suggest the intrinsic synergistic motion of go-to-grasp actions, offering a practical path, in hand rehabilitation contexts, for timing hand assistance in synergy with arm transport and with minimal setup burden.

More at link.

New upper-limb exoskeleton adapts to stroke patients in real time

 The hype is undeserved until tested in stroke survivors.

New upper-limb exoskeleton adapts to stroke patients in real time

A new five-joint exoskeleton senses stroke patients’ intent, adapting help in real time to speed upper-limb recovery.A Chinese research team has built a five-joint robotic exoskeleton that reads a patient’s movement intention and adapts its assistance on the fly, offering a more natural path to arm recovery after stroke and pointing toward safer, smarter neurorehabilitation. (CREDIT: Shutterstock)

Stroke can change your life in a single day. Around the world, more than 15 million people have a stroke each year, and roughly three quarters of them are left with long-term disability. If you are one of them, lifting a cup, buttoning a shirt or reaching for a shelf can become a daily test of patience. That reality is driving scientists to look for smarter, more intensive ways to retrain the brain and the body.

One of the most promising ideas pairs your nerves and muscles with robotics. Instead of asking you to fight through endless repetitions alone, these systems aim to guide your arm, measure your effort and quietly adjust the challenge as you improve.

A research team in China has now taken that concept a step further. Led by Professor Zeng-Guang Hou at the State Key Laboratory of Multimodal Artificial Intelligence Systems, they have built an upper-limb exoskeleton called CASIA-EXO and designed a control strategy that keeps you inside the loop of every movement. Their work, published in the IEEE/CAA Journal of Automatica Sinica, tries to give the robot something rehab often lacks: the ability to sense what you are trying to do and respond in real time.

The upper-limb exoskeleton robot for post-stroke rehabilitation; (b) Basic kinematic structure for 5-DOF mechanism, θ1,θ2,θ3 depict the adduction-abduction, internal-external rotation, flexion-extension of shoulder joint, θ4 depicts the flexion-extension of elbow joint, and θ5 depicts the supination-pronation of wrist joint. (CREDIT: IEEE/CAA Journal of Automatica Sinica)

Stroke Rehabilitation’s Long Road

After a stroke, your brain needs to relearn how to move the arm and hand. That takes thousands of repetitions and steady feedback. Traditional therapy can help; however, therapists have limited time, and sessions may not always be intense or frequent enough to unlock the full potential of neural plasticity.

Over the last decade, exoskeleton-style rehab robots have started to fill that gap. Systems such as ArmeoPower, ANYexo and EXO-UL8 let your shoulder, elbow and wrist move through multi-joint exercises while motors assist. They use control methods to encourage you to contribute rather than remain passive.

Still, most devices fall short in three key areas at the same time: detecting your movement intention, generating a realistic path for the arm and tailoring the level of assistance to your ability. Without all three, the robot can slip into either dragging your arm or leaving you under-supported.

A Robotic Suit That Moves Like An Arm

CASIA-EXO was designed to feel closer to a real limb. It has five degrees of freedom; three joints arranged at slanting angles to capture complex shoulder motion and two joints lined up in a chain for elbow and wrist. The metal and motors sit outside your arm, but the geometry matches the way your own bones move.

“CASIA-EXO is a five degree-of-freedom biomimetic exoskeleton that comprises three rotational joints adopting an oblique arrangement and two rotational joints co-locating in a serial chain,” Prof. Hou explains. In simple terms, the team built the robot to follow the same arcs and pivots that your joints naturally use.

Overview of the patient-in-the-loop control strategy. (CREDIT: IEEE/CAA Journal of Automatica Sinica)

To control such a device, engineers first built a detailed model of its dynamics. They examined how forces travel through the shoulder, elbow and wrist sections, then simplified that math so the robot can estimate unknown parameters as it runs. That lets the system adjust to different arm sizes, muscle strengths and movement styles.

Teaching the Robot to Read Intention

The real innovation lies in how CASIA-EXO decides what to do next. Instead of forcing your arm to follow a fixed script, the control design treats you as a partner in a loop.

“It consists of the intention-based trajectory planning and performance-based intervention adaptation,” says Chen Wang, a researcher on the project. The first part focuses on what you seem to want to do. An oscillator-based intention estimator tracks your timing and effort and folds in known patterns from normal human motion. With that information, it builds a multi-joint path for your arm that looks and feels like a natural reach.

The second part watches how well you keep up. A performance-based adaptive algorithm raises or lowers the assistance and resistance during training. If you start to handle the movement more easily, the robot backs off and lets your muscles do more work. If your arm trembles or stalls, the exoskeleton quietly adds support.

In practice, that means CASIA-EXO is never locked into one setting. It constantly reshapes both the path and the level of help to match your changing condition, second by second.

Real-time tracking performance in the upper extremity movements for a representative subject, where (x(t),y(t),z(t)) and (xr(t)yr(t)zr(t)) are the actual trajectory and generated trajectory in the workspace. (CREDIT: IEEE/CAA Journal of Automatica Sinica)

Putting the System to the Test

To see whether this approach works, the team first tried it on 10 healthy volunteers. Each person sat in a chair with their dominant arm strapped into CASIA-EXO. On a virtual reality screen, they saw wooden boxes they needed to move across a digital workspace and then bring back. The exoskeleton guided their real arm as they performed the task.

During these sessions, the system had to do three things at once. It needed to sense subtle hints of intention at the start of each reach, generate a smooth, joint-level trajectory and adjust the force field around the arm based on performance.

The results showed that the control strategy could steadily personalize both the movement and the level of intervention for each subject. Trajectories shifted as people sped up, slowed down or changed style. Assistance and resistance levels rose or fell when the volunteers improved or struggled. The loop between intention detection and assistance adaptation stayed stable throughout practice.

Although the tests involved healthy arms rather than stroke survivors, they provided an important proof that the algorithms and hardware can cooperate without fighting the user’s motion.

What Makes This Approach Different

Many rehab robots are either too stiff or too passive. They might move you along the same path regardless of your effort, or they might simply add strength without caring about timing. CASIA-EXO tries to avoid both extremes by weaving your intent into every decision.

The adaptation of robotic assistance and resistance accompanies the training performance for a representative subject, where E is the tracking error, Er and Ea are the error boundaries for mode switching, Fa and Fr are the assistance and resistance intervention. (CREDIT: IEEE/CAA Journal of Automatica Sinica)

Because trajectory planning and assistance tuning are tied together, the device can support motor relearning in a way that feels more like daily life. The exoskeleton does not just swing your elbow; it coordinates shoulder, elbow and wrist as a unit, closer to how you would actually reach for a cup or pull open a drawer.

This close cooperation between human and machine is what the researchers call patient-in-the-loop neurorehabilitation. Instead of leaving you outside the control system, the robot treats your nervous system as a key part of it. Over time, that kind of interaction may help your brain rebuild pathways for movement in a more natural rhythm.

The team believes this strategy can make robot-aided rehabilitation safer and more personalized. Once trials with stroke survivors begin, the next test will be whether it can also make recovery faster and more complete.

Practical Implications of the Research

If clinical studies confirm these early results, CASIA-EXO and similar systems could reshape stroke rehab. For survivors, a patient-in-the-loop exoskeleton may provide longer, more intensive training sessions without exhausting therapists, while still keeping your own effort at the center of every movement.

By sensing intention and adjusting support on the fly, the device can challenge you as you heal instead of locking you into a fixed difficulty level. That may speed up motor relearning, deepen neural plasticity and help you regain everyday skills like reaching, grasping and lifting.

For clinicians, tools like this could offer objective data about how your arm responds to therapy over time. That information might guide personalized rehab plans and reveal which control strategies work best for different levels of impairment.

In the long term, the same ideas could extend beyond stroke. People recovering from brain injury, spinal cord damage or orthopedic surgery might benefit from exoskeletons that adapt to their intentions and abilities.

As robotics and artificial intelligence advance, this research points toward a future where rehabilitation feels less like being moved by a machine and more like having a responsive partner helping you move yourself.

Research findings are available online in the journal IEEE/CAA Journal of Automatica Sinica.

Video mirror feedback induces more extensive brain activation compared to the mirror box: an fNIRS study in healthy adults

 And you somehow think stroke survivors will be able to easily get this and it works vastly better than just mirror therapy? What alternate universe do you live in? Testing in healthy adults? What fucking stupidity? I'd fire all of you!

Video mirror feedback induces more extensive brain activation compared to the mirror box: an fNIRS study in healthy adults

• Julien Bonnal,
• Canan Ozsancak,
• Fabrice Prieur &
• Pascal Auzou 

Journal of NeuroEngineering and Rehabilitation volume 21, Article number: 78 (2024) Cite this article

• Metrics details

Abstract

Background

Mirror therapy (MT) has been shown to be effective for motor recovery of the upper limb after a stroke. The cerebral mechanisms of mirror therapy involve the precuneus, premotor cortex and primary motor cortex. Activation of the precuneus could be a marker of this effectiveness. MT has some limitations and video therapy (VT) tools are being developed to optimise MT. While the clinical superiority of these new tools remains to be demonstrated, comparing the cerebral mechanisms of these different modalities will provide a better understanding of the related neuroplasticity mechanisms.

Methods

Thirty-three right-handed healthy individuals were included in this study. Participants were equipped with a near-infrared spectroscopy headset covering the precuneus, the premotor cortex and the primary motor cortex of each hemisphere. Each participant performed 3 tasks: a MT task (right hand movement and left visual feedback), a VT task (left visual feedback only) and a control task (right hand movement only). Perception of illusion was rated for MT and VT by asking participants to rate the intensity using a visual analogue scale. The aim of this study was to compare brain activation during MT and VT. We also evaluated the correlation between the precuneus activation and the illusion quality of the visual mirrored feedback.

Results

We found a greater activation of the precuneus contralateral to the visual feedback during VT than during MT. We also showed that activation of primary motor cortex and premotor cortex contralateral to visual feedback was more extensive in VT than in MT. Illusion perception was not correlated with precuneus activation.

Conclusion

VT led to greater activation of a parieto-frontal network than MT. This could result from a greater focus on visual feedback and a reduction in interhemispheric inhibition in VT because of the absence of an associated motor task. These results suggest that VT could promote neuroplasticity mechanisms in people with brain lesions more efficiently than MT.

Clinical trial registration

NCT04738851.

Introduction

Mirror therapy (MT) is commonly used for stroke rehabilitation. This technique consists of using the reflection in a mirror of the movements of a healthy limb to give the illusion of movement of the pathological limb. First proposed for phantom limb pain [1], MT was then used for motor rehabilitation of the post-stroke hemiparetic upper limb [2]. Recent meta-analyses have reported a beneficial effect of MT on upper limb motor recovery after stroke [3, 4].

Despite its effectiveness, the use of MT may be limited by difficulty with positioning for individuals with postural deficits, the need for bilateral training, or associated disorders such as aphasia or hemispatial neglect [5, 6]. New MT tools using virtual reality have been developed to improve the technique [7]. In this study, we focused on video therapy (VT) in which the mirror is replaced by a digital screen [8,9,10]. The use of these recent tools has been found to be feasible [11]. To our knowledge, there is no evidence of clinical superiority of VT over MT. The relatively high cost of these technologies makes it necessary to determine if they are indeed more effective than simpler, lower cost tools [7]. As such, it seems relevant to compare brain activation patterns between both modalities (MT and VT).

Many studies have explored the brain mechanisms of MT in both people after stroke and healthy individuals. MT activates the motor cortex, in particular the primary motor cortex (M1), premotor cortex (PMC) [12,13,14,15] and the precuneus (PC) [16,17,18,19] contralaterally to the side of visual feedback. In this study we focused more specifically on the activation of the PC as a determining factor of the effectiveness of the technique. Indeed, it has been shown that motor recovery following MT is correlated with PC activation [19]. One of the roles of the PC is to integrate the visual information from the environment and its transmission to the motor cortex to create a body self-perception [20]. Therefore, in MT the PC could be activated when the visual feedback gives the illusion of ownership of the visualized limb. It then seems relevant to assess the correlation between this activation and the quality of perception of the illusion.

Among the studies evaluating brain activity during MT, some used a real mirror [12, 13, 15, 17] and others a VT tool [14, 16, 18, 19], often for reasons of compatibility with the imaging method. To our knowledge, no study has directly compared the brain activation profiles of these 2 techniques. It seems appropriate to study these mechanisms in healthy subjects as a first step, in order to provide a rationale for future studies in patients. The literature on MT has shown similar activation patterns between healthy subjects [13, 16] and stroke subjects [14, 19]. A MT study conducted in healthy and stroke subjects found precuneus activation in both populations [21].

We chose to use fNIRS to determine the amount of activation of the cerebral regions of interest. This technique enables the evaluation of neurovascular coupling by measuring changes in both oxyhemoglobin (HbO₂) and deoxyhemoglobin (HbR) in the cortex. The portability of the fNIRS device means it can be used in the real-life environment, including to determine the cerebral mechanisms involved in rehabilitation [12, 13, 16, 19].

The first aim of the study was to compare cerebral activation (PC, PMC and M1) induced by MT and VT tasks using functional near infrared spectroscopy (fNIRS). We hypothesized that VT would lead to greater activation of each region. The second aim of this study was to evaluate the correlation between individuals’ perceptions of the illusion of movement for the two mirrored feedback modalities (MT and VT) and brain activation. We hypothesised that the stronger the illusion of movement, the greater the activation of the PC.

More at link.