When refering to evidence in academic writing, you should always try to reference the primary (original) source. That is usually the journal article where the information was first stated. In most cases Physiopedia articles are a secondary source and so should not be used as references. Physiopedia articles are best used to find the original sources of information (see the references list at the bottom of the article).
If you believe that this Physiopedia article is the primary source for the information you are refering to, you can use the button below to access a related citation statement.
Robotic upper limb rehabilitation uses powered or passive devices to guide arm and hand movement, supplementing therapist-led practice by increasing training dose, repetition and consistency.
End-effector robots attach at a single distal point for quick set-up, whereas exoskeletons align with anatomical joints for precise multi-joint control but cost more.
Stroke has the strongest evidence, with robot-assisted arm training improving activities of daily living, arm function and strength when added to conventional therapy.
Sessions of 30 to 60 minutes, several times weekly, have evidential support, with progression through assistive, correction and resistance modes as control improves.
Robotic upper limb rehabilitation uses powered or passive mechanical devices to assist, resist or guide movement of the shoulder, elbow, wrist and hand during therapy. It supplements, rather than replaces, therapist-led practice and is used primarily to increase the dose, repetition and consistency of movement training beyond what is feasible with manual therapy alone.[1]
The rationale for robotic rehabilitation rests on several established principles:
Repetition-dependent plasticity: Repeated practice of a movement drives cortical reorganisation and motor relearning.[1][2]
Consistency and objectivity: Robotic systems deliver standardised, measurable training and can record quantitative data in a way manual therapy cannot easily replicate.[2]
Labour and resource limitations of manual therapy: Therapist availability, cost and physical demand limit how much high-intensity, high-repetition therapy can realistically be delivered by hand, driving interest in robotic and hybrid technologies.[1]
Upper limb robotic devices fall into two broad structural categories, end-effector and exoskeleton, which differ in how the device attaches to and controls the patient’s limb.[2][3] This page also covers several other technologies commonly used alongside or built upon these two structures, namely soft or wearable robotics,[4] smart textiles,[5] hybrid robotic and functional electrical stimulation (FES) systems,[6] and virtual reality (VR), augmented reality (AR) and extended-reality-integrated systems.[7] All six are described in detail below.
Device descriptions throughout this page refer to a device’s number of “degrees of freedom” (DOF): the number of independent directions in which a joint or limb segment can move.[14] Understanding this helps a clinician judge, how closely a device’s movement capability matches the joint or joints being targeted in treatment.
DOF between joins varies due to joint type and structure:
Shoulder: three DOF, comprising flexion/extension, abduction/adduction, and internal/external rotation.
Wrist: two DOF, flexion/extension and radial/ulnar deviation, plus pronation/supination contributed by the forearm at the distal radioulnar joint.
Hand and fingers: a substantially greater number of DOF across the carpal, metacarpophalangeal and interphalangeal joints, reflecting the complexity of independent finger and thumb movement.
A device’s DOF count indicates how many of these movements it can support or train concurrently. Low-DOF devices, target a single movement pattern, whereas devices with six or more DOF, support coordinated multi-joint movement approximating natural arm function.[14]
End-Effector Robots
Mechanics of the robot
The patient interacts with the robot at a single distal point, such as a handle or forearm cradle, while the proximal segments of the limb (Shoulder and Elbow) remain unsupported by the device.[3][15]
Advantages
Quick to set up in a clinic session and easy to adapt to patients of different limb lengths and sizes, making them practical for use across a caseload with varied presentations.[2][3]
Disadvantages
Limited control of proximal joints, which can allow compensatory movement strategies to develop during training.[3]
Clinical Suitability
Shoulder and elbow retraining after stroke (a recent literature review specifically examined end-effector systems for this purpose[15]) and for patients needing rapid set-up across variable body sizes.
Examples
There are many examples of End-Effector Robots, which include:[2]
MIT-MANUS: a two DOF device using a five-bar linkage to guide shoulder and elbow movement.
MIME: based on the PUMA 560 industrial arm, which straps the affected limb and drives mirrored movement of the healthy side
ARMassist: a low-cost, portable device with optical and force sensors for home telerehabilitation
Bi-Manu-Track: for forearm and wrist training with adjustable intensity, speed and resistance.
ReoGo: a three-dimensional shoulder and elbow device.
NeReBot and GENTLE: Suspension or cable-driven variants, which use cables to provide gravity-compensated, low-inertia movement assistance.
Amadeo Robot: a five DOF device, providing motion to one or all five fingers through a passive rotational joint placed between the fingertips and an entity that moves (the thumb has two passive rotational joints).[16]
The table below shows some of these device's setup and use in practice:
Demonstration of another suspension driven end-effector type robot
Exoskeleton Robots
Mechanics
The device’s joints align with the patient’s own anatomical joints and attach at multiple points along the upper limb. This allows the therapist to target individual joints directly and reduce unwanted compensatory movements.[2][3]
Advantages
More precise isolated-joint training and improved multi-joint coordination, useful where a specific joint needs to be isolated within a treatment plan.[3]
Disadvantages
Greater design complexity, longer fitting time, higher purchase cost, and less flexibility to move quickly between patients of different sizes.[3]
Clinical Suitability
Multi-joint or proximal control training,[21] and paediatric use in CP where growth requires an adjustable, joint-specific fit.[13]
Examples
Exoskeleton Robotics have different joint-driven movement, examples of robots falls into 3 categories;
Harmony: a five DOF bilateral exoskeleton.[2] The following video demonstrates Bilateral upper extremity robotic rehabilitation with the Harmony system.
AGREE exoskeleton: evaluated in a single-blinded randomised controlled trial for restoring arm function after stroke.[21]
Pneumatic-driven examples:
Pneu-WREX and its prototype T-WREX, and BONES: which enables full shoulder rotation using parallel cylinders.[2]
Robotic-assisted Upper Extremity Repetitive Trainer (RUPERT): which pairs pneumatic artificial muscles with FES to enable active grasping in patients with high flexor tone.[2]
Elastic actuator-driven examples:
ANYexo: a seven (or nine) DOF exoskeleton using series elastic actuators for compliant, torque-controllable interaction;
NeuroExos Shoulder-Elbow Module Platform (NESM): for shoulder and elbow rehabilitation;
Pneumatic, tendon-driven or twisted-string actuators assist finger flexion and extension without a rigid frame, allowing the hand to remain relatively free for functional tasks.[24]
Advantages
Lightweight and comfortable for patients to wear, and well suited to home use and functional task practice between clinic visits.[24][25]
Disadvantages
Less precise force and range-of-motion control than rigid systems.[24]
Clinical Suitability
Hand and grip-focused rehabilitation in subacute to chronic stroke, and in cervical SCI, where loss of hand function is a high rehabilitation priority and specialist in-person therapy is often difficult to sustain after discharge.
A systematic review and meta-analysis found soft robotic gloves improved hand outcomes in stroke patients,[24] including with independent home use,[25] and a pilot study of a home-based soft robotic glove in people with chronic cervical SCI found improvements in hand function with self-administered use over 12 weeks.[26] A more recent review has specifically examined the combination of soft robotic gloves with functional electrical stimulation (FES) for hand rehabilitation after stroke.[27]
Examples
These are modular, fabric-based soft gloves that uses twisted-string actuators with flexible structures,[4] developed specifically for assistance and rehabilitation use.[28]
The following table contains examples and videos that demonstrate these gloves in use:
Sensor-based glove supporting hand movement recovery in stroke patients
Smart Textiles
Mechanics
Knitted or woven fabric integrates sensors and electrodes, avoiding rigid hardware and conventional gel electrodes and allowing patients to wear the garment comfortably for extended periods.[32] An increasing proportion of these systems incorporate biofeedback sensors that measure muscle activity or movement quality during exercise, giving the patient and clinician real-time performance information.[5]
Applications
Surface EMG biofeedback sleeves for self-administered training, developed and evaluated through a co-design process with stroke survivors and clinicians.[33] E-textile electrodes as an alternative to hydrogel pads for delivering electrical stimulation, tested for use in stroke rehabilitation.[34]
Clinical Suitability
Patients needing home-based, self-monitored biofeedback training, or those with skin sensitivity to conventional electrodes.[32]
Examples
Smart textile gloves are in the development phase with biomedical engineering studies validating the use of these textiles in rehabilitation.[33]
The following video demonstrates a Polymer based tendon driven wearable system:
Combines robot-guided movement with functional or neuromuscular electrical stimulation (FES/NMES), so the device provides external support to the limb while stimulation activates the patient’s own muscles, useful where a patient has some voluntary muscle activity but not enough to complete a movement independently.[6]
The following video shows NMES and FES incorporated exoskeleton:
A narrative review of nine hybrid FES-robot systems, searching IEEE Xplore, Scopus and PubMed up to October 2022, classified them by actuator type:[6]
DC motor-driven (four systems): simple and back-drivable, meaning the patient can move the device with relatively little resistance, including a two DOF NMES-robot for elbow and wrist extension and a seven DOF exoskeleton combining FES with shoulder, elbow and wrist support.
Pneumatic-driven (two systems): capable of producing greater assistive force, including RUPERT with eight-channel FES for reach-to-grasp training
Passive, spring-based (three systems): require no electrical power source and adapt readily to individual patient size, including the RETRAINER-Arm, evaluated in a 72-patient randomised controlled trial.
Clinical Suitability
Patients with minimal to emerging voluntary movement who need augmented muscle activation alongside guided motion, and increasingly, patients who could benefit from real-time, fatigue-adaptive stimulation dosing.
VR, AR and Extended-Reality-Integrated Systems
Mechanics
Movement training is paired with visual feedback ranging from non-immersive screens to fully immersive headsets, often gamified to increase engagement and repetitions.[7]
Video using VR and NMS for upper limb rehabilitation:
Improved engagement and adherence, adjustable task difficulty and potential for remote delivery.
Evidence base
A systematic review and meta-analysis found virtual, augmented and mixed reality (VAMR) therapy produced greater upper limb and activities-of-daily-living gains than conventional rehabilitation,[38] and haptic glove systems combined with semi-immersive VR have shown benefit for upper extremity motor recovery after stroke.[39]
Examples
Collaborative robot plus AR example: A UR5e collaborative robot was paired with a Microsoft HoloLens 2 augmented reality headset to create a combined haptic and visual gamified rehabilitation interface. Two control modes, an autonomous predefined-path mode and a velocity-controlled joystick mode able to provide either guidance or resistance.[40]
Robotic gamified framework example: A separate framework built around a UR10e collaborative robot with admittance control; synchronising eight-channel sEMG data with three gamified reaching tasks (Odyssey, Skyward Stride and Kora Game), exporting kinematic and EMG datasets for the therapist.[41]
Clinical Suitability
Patients across severity levels where motivation or engagement is a limiting factor, and settings where synchronised kinematic and EMG data would help guide therapy decisions.
Key Evidence for Robotic Rehabilitation by Condition
Stroke
Robotic-assisted rehabilitation in stroke aims to recover function in the hemiparetic arm. Key evidence includes:
A network meta-analysis found that robot-assisted arm training improves activities of daily living, arm function and arm muscle strength when added to conventional therapy.[8]
A systematic review and meta-analysis of 18 randomised controlled trials found that robotic arm interventions lasting 30 to 60 minutes per session produced significant improvements in upper limb function.[9]
Soft robotic gloves have shown significant gains in hand function,[24] including when used independently at home.[25]
Exoskeleton-based systems, such as the AGREE device, found to assist in restoring arm function after stroke in a randomised controlled design.[21]
Hybrid FES and robotic approaches, including AI-adaptive systems, are an emerging area specifically targeting upper limb recovery after stroke, though current evidence is limited to small pilot studies.[42][43]
A randomised controlled trial in acute stroke patients examined robot-assisted hand therapy using the Amadeo Robotic System by Tyromotion.[16] They found that comparable to traditional Occupational Therapy methods, patients within the robotic therapy group made significant improvements in Fugl-Meyer Scale (FM), and Box and Block Test (BB) at the end of treatment (4/5 weeks) and maintained improvements after a 3-month follow-up. This result is very important because the gain achieved is not exercise or time-dependent, but could be secondary to the reorganisation of brain structures. [16] The Amadeo protocol combined continuous passive therapy, assisted movement therapy, and balloon training (active, target-oriented tasks in a virtual environment).[16]
Further evidence has demonstrated the clinical feasibility of MIT-MANUS, MIME, ARM-Guide, T-WREX and NeReBot.[44] When conventional therapy was matched with robotic therapy in duration and intensity, no statistically significant difference in Fugl-Meyer scores was observed between the two. When robotic therapy was added to conventional therapy, however, Fugl-Meyer scores improved significantly, with the same pattern observed for motor control on the Motor Status Scale, indicating that increased overall therapy dose, whether conventional or robotic, is associated with improved outcomes.[44]
Spinal Cord Injury (SCI)
Hand function is essential for activities of daily living, and cervical SCI, at levels above C7, presents with decreased elbow, wrist and hand function. A comprehensive review of robotic-assisted arm training in cervical SCI found the approach feasible and safe, with initial positive effects on arm function and movement quality as an adjunct to conventional therapy.[10] A study of the Armeo Spring system in subacute SCI demonstrated significant improvement in GRASSP sensibility scores among subjects with partial hand function at baseline,[22] while a systematic review and meta-analysis found only limited evidence that VR interventions improve upper limb motor recovery after SCI compared with conventional physical therapy.[45]
Multiple Sclerosis
An evidence-based review found robotic-assisted upper limb training clinically applicable in MS, although overall study quality remains variable and larger, high-quality trials are still needed.[11] Evidence has also shown that functional capacity tests significantly improved after treatment with the Armeo Spring, and improvements were maintained at 2-month follow up.[46] Research using the HapticMaster robot combined with VR showed positive effects on perceived function, and MS patients with upper limb function considered marked-to- severe showed considerable improvement in clinical tests.[47]
Parkinson's
Evidence remains limited to small pilot studies. A trial using the Bi-Manu-Track robot in a small sample of patients found significant improvements in the nine-hole peg test and the upper limb Fugl-Meyer score, sustained at follow-up.[12]
Cerebral Palsy (Paediatric)
A systematic review of robot-assisted therapy in children with CP, using both exoskeletons and end-effectors, found improvements in upper limb movement and manual dexterity, though study designs remain heterogeneous.[13] See also page on Paediatric Robotic Rehabilitation
Treatment Protocol for Robotics Use
Assessment
Before applying robotics for upper limb rehabilitation it is important to have a thorough initial assessment to:
Robotic technology has become a valuable adjunct to conventional upper limb rehabilitation, offering repetition and measurable feedback that manual therapy alone often cannot sustain.[1] Devices fall into six broad categories, end-effector systems, exoskeletons, soft or wearable robotics, smart textiles, hybrid FES-robotic systems, and VR/AR-integrated systems. Each device type is suited to different needs: end-effectors offer speed and simplicity,[2] exoskeletons allow precise multi-joint control,[3] soft robotics and textiles extend training into the home,[24] and hybrid FES or VR/AR systems support engagement and residual movement in more complex cases.[6][7]
Evidence is strongest for stroke, where robot-assisted training reliably improves arm function, strength and activities of daily living alongside standard care.[8] Support is growing but still developing for SCI,[10] MS,[11] PD[12] and paediatric CP.[13]
Successful use depends on thorough assessment, device selection matched to patient severity and goals, appropriate dosing (typically 30 to 60 minute sessions several times weekly), and ongoing attention to safety, including screening for contraindications and monitoring for compensatory movement or fatigue.[2][48] Robotics should always complement, not replace, skilled therapist-led practice. As the evidence base matures, particularly for adaptive and AI-integrated systems, robotic rehabilitation is likely to play an increasingly central role in upper limb recovery.
↑ 7.07.17.2Isaac LD, V. M, Soms N. Chapter 11 - Augmented and virtual reality in medical robotics. In: Musiolik TH, Kannan H, Rodriguez RV, Rege M, eds. AI-Powered Developments in Medical Robotics. Medical Robotics and Computer Assisted Surgery: AI-enhanced, Data-driven, and Evidence-based Approaches. Academic Press; 2026:287-321.
↑ Neuroengineering Lab. Merging neural stimulation and exoskeletons to enhance hand functions in neurological patients. Available from: https://www.youtube.com/watch?v=Cyc6_YTq-Q0 [last accessed 16/8/2026]
↑ Neuroengineering Lab. Immersive virtual reality with synchronous neurostimulation for upper-limb recovery after stroke. Available from: https://www.youtube.com/watch?v=YQTYrCTsvPI [last accessed 16/8/2026]
↑ 44.044.1Norouzi-Gheidari N, Archambault PS, Fung J. Effects of robot-assisted therapy on stroke rehabilitation in upper limbs: systematic review and meta-analysis of the literature. J Rehabil Res Dev. 2012;49(4):479–96.
