# Tendon-driven hand assistance: consolidated prior-art comparison

Reviewed: 2026-10-15. Scope: merge the user's supplied comparison with the earlier research list; verify distinctive claims using primary papers, university repositories and original project repositories. This is an engineering literature note, not a systematic review or a prescription for patient forces.

## Current proposed device

Four long fingers only (index, middle, ring, little); no thumb, wrist or elbow actuation. Eight motors: separate flexion and extension actuation for each finger. Fingertip caps anchor tendons, flexible clip-on rings guide them, and a forearm support holds the motors. Desired modes: loose/rest, armed slack take-up, and assisted movement. ESP32 physical switches must work without the iPhone; the companion app uses Wi-Fi and camera-based motion calibration. These are user-selected requirements, not features already validated on hardware.

Force adaptation requires actual tension sensing and bounded local control. Camera landmarks alone do not measure tendon force, interface pressure or a patient's permissible assistance. Exact assisted daily task, motor sizing and acceptable patient group remain unresolved.

## How the two lists compare

The user's list adds explicit Exo-Glove Poly II, MGlove-TS (2024), and the biomimetic single-motor paper (2023). It overlaps with HERO, Bagneschi's open-ring/thimble design, and FLEXotendon. The earlier list adds the separate Robotica modular-glove paper (2025), TAKTO ONE, and the ESP32 open-source project. Park's 2020 surface-mounted-actuator paper is retained as an additional mechanical reference. These make ten distinct records; related papers should not be merged into one device or one clinical evidence claim.

| ID / reference | Actuation, attachments and mounting | Fit to our four-finger concept | Evidence and limits |
|---|---|---|---|
| P01 — [HERO Grip Glove, 2020](https://doi.org/10.1186/s12984-020-00659-5) | Two linear actuators; dorsal extension and palmar flexion tendons; cable-tie guides; open-palm interface and wrist brace. | Strong stroke/donning reference; buttons support operation independently of a smartphone. | Eleven stroke survivors. Reported mean grip 12.7 N and pinch 11.0 N are task outputs, not opening-force prescriptions. No independent eight-motor control. |
| P02 — [Bagneschi et al., 2023](https://doi.org/10.1109/TOH.2023.3273908) | TPU open rings and fingertip thimbles, embedded PTFE guides, releasable screw terminals; opposing tendon paths driven by one multi-groove pulley on the hand dorsum. | Closest finger-interface precedent. Rings and caps can guide our CAD study. | Prototype actuates thumb, index and middle together; not four independent long fingers. Unloaded open rings ease clamping, but do not establish zero resistance or a validated automatic rest state. |
| P03 — [Exo-Glove Poly II, 2019](https://pubmed.ncbi.nlm.nih.gov/30566026/) | Polymer wearable interface; optimized antagonistic tendon routing with single-actuator coupling; passive thumb opposition; remote drive unit. | Important reference for fit, routing and reduced actuator count. | Two SCI participants; 104 g glove and 1.14 kg actuation unit reported in the abstract. SCI performance must not be transferred to stroke without evaluation. |
| P04 — [MGlove-TS, 2024](https://doi.org/10.1016/j.mechatronics.2024.103141) | Modular flexible glove; forearm twisted-string actuators; routing rings and fingertip locking pins. Original configuration uses elastic return for extension. | Strong modular mounting and tendon-adjustment reference. | Index fingertip outputs depend on actuator length; approximately 15.82–17.95 N in Table 3. TSA pulling force is a separate measurement. Preliminary evaluation, not a general stroke force requirement. |
| P05 — [FLEXotendon Glove-III, 2023](https://pubmed.ncbi.nlm.nih.gov/37124112/) | Tendon tension sensing, fingertip force sensing, quick connections, smartphone calibration/operation and voice control. | Strongest app-plus-measured-force architecture among these records. | Five-DoF device, case study with two SCI participants. Not proof of our eight-motor routing, iPhone/Wi-Fi stack or camera-based force estimation. |
| P06 — [Biomimetic tendon mechanism, 2023](https://doi.org/10.3390/s23042272) | Single motor coordinates flexion/extension; multiple tendon paths, guidance beads/sheaths and mechanical routing stops. | Strong reference for opening all finger joints rather than only pulling the fingertip. | Artificial finger, stiff artificial-finger condition and nine healthy participants. Simulated stiffness is not a stroke-patient trial. |
| P07 — [Modular tendon-actuated glove, Robotica 2025](https://doi.org/10.1017/S0263574725102464) | Separate publication from P04. Ring-based modules, forearm actuation and TSA arrangements; investigates active flexion/extension and coupled abduction/adduction. | Useful tendon-length simulations and routing geometry; Figures 3–6 show configurations. | Kinematic simulations and preliminary user tests. A configuration explored in simulation is not evidence every motion was clinically validated. |
| O01 — [TAKTO ONE](https://github.com/molanocortes/takto-one) | Four instrumented long-finger assemblies, eight planned servos for those digits, magnetic joint encoders, CAD, firmware and browser twin. | Closest open-source four-finger/eight-servo design allocation. | Author-reported prototype. Repository explicitly says force rendering, transparency and assistance are not characterized by measurements. Verify actual fitted hardware; designs and demos evolve. |
| O02 — [ESP32 hand_exoskeleton](https://github.com/MengyangGao/hand_exoskeleton) | ESP32, seven SC09 bus servos; linkages for the four long fingers and tendon transmission for the thumb. | Useful controller/software reference. | Hybrid research/student project; not the same all-tendon cap-and-ring mechanism. Repository performance claims are not clinical validation. |
| P08 — [Park et al., 2020](https://doi.org/10.3389/fbioe.2020.548947) | Surface-mounted motor modules, paired tendons and a two-layer glove; some fingers have two motors for stiffness control. | Useful spool/actuator packaging and opposing-tendon coordination. | Tested as a glove on a robotic hand skeleton, not a stroke-worn glove. Does not validate our exact two-independent-pullers-per-finger control. |

## Corrections to the supplied summary

1. **MGlove-TS and Robotica are separate papers.** P04 is Mechatronics 98 (2024), 103141; P07 is Robotica 43 (2025), 3902–3920. P07 explicitly describes the earlier MGlove design as using passive elastic extension, then studies additional tendon configurations. Do not attach active bidirectional results from one configuration to the entire series. [P04](https://doi.org/10.1016/j.mechatronics.2024.103141), [P07](https://doi.org/10.1017/S0263574725102464)
2. **“About 10 N per finger” is not a universal requirement.** HERO lists at least 10 N extension per digit as a design specification; its 80 N shared actuator/tendon tension is not a measured 80 N finger force. The study says individual extension force was not measured. [P01](https://pmc.ncbi.nlm.nih.gov/articles/PMC7045638/)
3. **Output force and required opening force are different.** A blocked-tip test, pinch/grip measurement, cable pull, or motor stall rating does not establish the force needed to open a particular patient's hand.
4. **Separate motors still need slack control.** Winding one tendon while the other holds its length can constrain motion. Both channels must coordinate release/take-up. Simultaneous tension increases stiffness and potentially compressive loading; motor count alone does not solve this.
5. **Loose at rest is a behavior to verify.** P02's open-ring geometry is relevant, but we must measure residual tension, cap retention and spool behavior. A geared servo can resist motion with PWM disabled. Emergency release must work without powered unwinding.
6. **Camera-driven adaptation must use force feedback.** Vision may estimate movement and tracking error. A tension sensor supplies the load measurement; bounded ESP32 control decides whether to continue or stop. Occluded joints and fingertip caps can degrade tracking. No reviewed source validates our proposed MediaPipe-only force estimate.
7. **Avoid broad safety generalizations.** The supplied assertion that almost all designs have mechanical stops and clinical validation is too strong: some use routing stops, some report small healthy-subject or robotic-hand evaluations. Record protections and test population per device.
8. **No visual files were supplied in the pasted comparison.** The “visual examples” headings were captions with empty spaces. Source figures are linked here; they are not represented as locally downloaded images.

## Force figures worth retaining

| Record | Quantity / context | Interpretation |
|---|---|---|
| P01 | Grip 12.7 ± 8.9 N; pinch 11.0 ± 4.8 N | Mean assisted task forces, not individual finger-extension forces. |
| P01 | 80 N shared actuator/tendon tension | Do not divide by five and call that measured force at every fingertip. |
| P04 | Index fingertip: 15.82 ± 1.28 N at 41 mm TSA length; 17.95 ± 1.04 N at 53 mm | Measured device output in the paper's setup. Thumb values differ. |
| P04 | Up to 57 N unconstrained actuator force at 53 mm | Actuator pull, not the fingertip value above. |
| P06 | Stiff artificial-finger comparison | Evidence about routing/trajectory, not a patient-safe load threshold. |

P04 values: university-hosted paper, Table 3 and conclusion. The 41–53 mm figures describe the TSA module's twisted-zone length; **they are not 41–53 mm of available tendon stroke**. [University PDF](https://usiena-air.unisi.it/bitstream/11365/1255604/1/1-s2.0-S0957415824000060-main%20%281%29.pdf)

## Recommended starting points for our design

These are engineering judgments drawn from the comparison, not rankings reported by the papers.

| Workstream | Start with | Reason |
|---|---|---|
| Fingertip cap and flexible ring CAD | P02, then P04/P07 | Open attachments, guide placement, adjustable anchoring and modularity. |
| Extension routing | P06 | Investigates coupled joint motion and routing stops. |
| Stroke usability and physical controls | P01 | Patient-involved development and an open interface for clenched hands. |
| Measured-force control and app workflow | P05 | Distinguishes tendon load, fingertip contact and user interface. |
| Four-finger/eight-servo packaging | O01 | Reusable design reference, subject to prototype limitations and licensing. |
| ESP32 integration | O02 | Concrete embedded project; mechanics require redesign for our concept. |
| Tendon travel simulation | P07 | Model cable travel before selecting spool radius and servo range. |

Minimum bench study before a worn prototype: one artificial finger; measure each cable's tension and travel through the real guides; verify one channel releases while the other pulls; test stop, switch-only use, stale commands, sensor failure and app loss. Extend to four fingers only after those behaviors work. Simulations supplement these measurements and do not certify patient safety.

## Files and access status

- `sources.json`: machine-readable records, DOI/source/PDF/code links and access limitations.
- `references.bib`: citation records for all ten references.
- `USER_COMPARISON.md`: preserved user-supplied summary, labeled as supplied rather than verified evidence.
- `fetch_papers.py`: download/retry available full-text URLs, validating PDF headers and recording hashes/status.
- `papers/`: Four full-text PDFs saved after network access was restored: HERO 2020, biomimetic tendon 2023, modular glove 2025 and Park 2020. SHA-256 and retrieved URLs are recorded in sources.json. Three PDF links failed (403 or non-PDF content); see download-report.json.

No third-party CAD, code or figures were copied. Repository links identify where to inspect them. Check each item's license before incorporating assets into our project; an accessible paper does not necessarily provide reusable CAD files.