PaperlayerView exact revision on arXiv

Source-linked AI summary

PalmSpace: Towards a Versatile On-Palm Interaction Space through Unified Touch Modeling

Chentao Li, Mingze Gao, Runze Sun, Zhaoguo Wang, Jianjiang Feng, Jie Zhou

arXiv:2610.10370v1cs.HCcs.CV

TL;DR

The bare palm has tactile advantages but lacks an explicit coordinate system and embedded touch sensing, while prior systems often represent only parts of palm interaction. PalmSpace combines wrist-worn infrared sensing with a unified representation of contact, location, and mode; it reports localization, contact-detection, and interaction-state results, with prior palm approaches still bounded by discrete regions or setup constraints.

  • Problem

    The bare palm lacks an explicit coordinate system and embedded touch sensing, and prior on-palm systems typically expose only one layer of input semantics.

    Source: §1 · p. 2

  • Method

    PalmSpace uses wrist-worn infrared sensing and a unified representation of contact state, palm location, and interaction mode.

    Source: §3.1 · p. 5, §4.1 · p. 7

  • Results

    6.7 mm mean localization error, 98.9% contact detection accuracy, and 96.7% F1-score for four-class interaction-state recognition were reported.

    Source: §1 · pp. 2–3

  • Takeaways & Limitations

    Evaluation covered absolute pointing, eyes-free symbolic input, and representative multi-finger gesture control.

    Source: §1 · p. 3

    PalmType was limited to predefined key regions, while PalmGesture required four-corner interaction-area setup and strict finger-orientation postures.

    Source: §2.2 · p. 4

Abstract

from arXiv · show

As smart glasses and lightweight MR devices become increasingly practical, input remains a key challenge. The bare palm is an always-available, tactile, and proprioceptively accessible surface, but it has neither an explicit coordinate system nor embedded touch sensing. Prior on-palm systems typically expose isolated touch events, discrete regions, continuous trajectories, or task-specific gestures, limiting the palm's ability to support precise selection and gesture manipulation through a common input representation. We present PalmSpace, a wrist-worn infrared system that exposes mode-aware, body-referenced absolute input on the bare palm without per-user sensing calibration. At the interaction level, PalmSpace jointly represents contact occurrence, interaction mode, and palm-referenced absolute location; at the model level, it learns these coupled outputs through a shared real-time representation. In leave-one-participant-out evaluation with 17 participants, PalmSpace achieved 6.7 mm mean localization error, 98.9% contact detection accuracy, and 96.7% F1 for four-class interaction-state recognition. User studies further demonstrated absolute pointing and dragging, eyes-free digit input, and representative multi-finger controls including scrolling and pinch-based map manipulation. These results show that a morphologically variable bare palm can function as a transferable, mode-aware interaction surface.

1 Introduction

PalmSpace addresses the lack of a common palm-input representation by jointly modeling contact, interaction mode, and absolute location. Its wrist-worn infrared system supports cross-user palm interaction, with reported localization, contact-detection, and interaction-state results and studies of pointing, digit entry, and gestures.

1 Introduction

  • The bare palm lacks an explicit coordinate system and embedded touch sensing, while prior systems often expose only isolated events, regions, positions, or task-specific gestures.The paper identifies a missing common representation that combines contact, body-referenced absolute location, and interaction mode.Source: §1 · p. 2
  • PalmSpace jointly represents contact occurrence, interaction mode, and palm-referenced absolute location through a shared multi-task representation.The system uses wrist-worn infrared sensing, palm-normalized coordinates, and a real-time multi-task Transformer without per-user sensing calibration.Source: §1 · p. 2
  • 6.7 mm mean localization error, 98.9% binary contact detection accuracy, and 96.7% F1-score for four-class interaction-state recognition were achieved in leave-one-participant-out evaluation.The evaluation used 17 participants.Source: §1 · pp. 2–3
  • User studies covered Fitts’s Law pointing, eyes-free digit entry and handwriting, list navigation, and map manipulation, alongside outdoor evaluations in daytime and low light.Representative multi-finger controls include scrolling and map manipulation.Source: §1 · pp. 2–3
  • PalmSpace is presented as turning a morphologically variable bare palm into a transferable, mode-aware interaction surface.The stated contribution includes absolute input without per-user sensing calibration.Source: §1 · p. 3

2 Related Work

Prior head-worn and on-hand input methods trade among precision, portability, sensing demands, and expressive control. PalmSpace is positioned against approaches that leave palm input fragmented or require calibration, bulky hardware, or limited interaction representations.

2.1 Input Methods for Head-Worn Interaction

  • Head-worn interaction methods involve trade-offs: gaze can bring fatigue and precision constraints, mid-air input has limited haptic feedback, and physical devices require dedicated hardware or stable external surfaces.These constraints motivate input that is portable and suitable for eyes-free daily use while supporting precise control and diverse commands.Source: §2.1 · p. 3

2.2 On-Hand Input Methods

  • Earlier palm interfaces used predefined key regions or calibrated continuous surfaces with posture constraints, leaving contact, absolute location, and interaction mode rarely modeled together.PalmType used discrete infrared proximity sensors; PalmGesture required users to define the interaction area by touching four corners and imposed strict finger-orientation constraints.Source: §2.2 · pp. 4–5
  • Prior palm systems remain constrained by bulky hardware, finger instrumentation, calibration, or viewpoint dependence, while many address only touch detection, localization, or specific gestures.The comparison spans sensing setup, calibration, localization, expressiveness, and evaluation scope; PalmSpace combines continuous absolute positioning with multi-finger mode recognition without per-user sensing calibration.Source: Table 1 · p. 3, §2.2 · p. 5

3 PalmSpace Overview

PalmSpace uses wrist-worn infrared sensing to capture bimanual finger-to-palm interaction, organizing the palm as a structured space for precise input and multi-finger control.

3.1 Proof-of-Concept Hardware

  • The wrist-worn infrared prototype captures fine-grained palm contact cues for touch detection, localization, and gesture recognition under stable imaging conditions.The prototype is wired at this stage.Source: §3.1 · p. 5

3.2 Versatile Interaction Modes

  • PalmSpace also supports palm-based Scroll and Pinch, structuring interaction by finger configuration and contact surface for symbolic input and manipulation.The paper describes the larger, smoother contact area as better supporting stable multi-finger control.Source: §3.2 · p. 6
  • Single-finger input includes Left Yaw, Vertical Touch, and Right Yaw postures, with freely adjustable pitch and contact on palm or finger surfaces.The postures are based on finger yaw relative to the palm.Source: §3.2 · p. 6

4 PalmSpace Method

PalmSpace unifies discrete touch semantics and continuous palm-relative location in a shared model, using normalized coordinates and jointly trained prediction tasks.

4.1 Unified Touch Modeling Framework

  • PalmSpace jointly models contact state, interaction mode, and palm-normalized absolute location in one real-time, end-to-end model without per-user sensing calibration.The shared representation couples discrete interaction semantics with continuous spatial input.Source: §4.1 · p. 7

4.2 Palm-Normalized Touch Representation

  • Palm-relative coordinates address hand-size variation that makes direct fingertip-coordinate regression sensitive to user anatomy.The dataset's smallest-to-largest hand gap reaches 4.5 cm.Source: §4.2 · p. 7
  • The system centers horizontal position at the wrist midline and scales it by palm breadth, while vertical position is referenced to the palm base and scaled by palm length.This maps the palm to lateral range [−0.5, 0.5] and longitudinal range [0, 1], supporting cross-user consistency and transfer across screen sizes.Source: §4.2 · p. 7

4.3 Multi-Task Transformer Architecture

  • A visual feature extractor feeds a lightweight Transformer and two task-specific heads for interaction-state classification and location regression.The model classifies Non-contact, Single-finger, Scroll, and Pinch; for multi-finger gestures, it regresses the index finger as the primary interaction point.Source: §4.3 · pp. 7–8

4.4 Training Objective

  • End-to-end training jointly optimizes four-state classification and masked mean-squared-error localization, excluding non-contact samples from spatial loss.The classification and regression loss weights are set to 1 and 100, respectively.Source: §4.4 · p. 8

5 Offline Evaluation

The offline evaluation uses a finger-to-palm interaction dataset with labels for contact, gesture, location, and palm size, and derives localization ground truth from an overhead camera.

5.1 Data Collection

  • The collected dataset covers the interaction variations shown in Fig. 3 and labels touch state, gesture category, hand-surface coordinates, and palm size.Gesture categories are Single, Scroll, and Pinch.Source: §5.1.1 · p. 9

5.1.1 Pipeline.

  • A wrist-mounted infrared camera supplies model input, while an overhead camera, AprilTag calibration area, homography, and tracked fingertip keypoints produce metric palm-plane ground truth.MediaPipe tracks the fingertip keypoints in the overhead view.Source: §5.1.1 · p. 9
  • The ground-truth pipeline is markerless at the hand, avoiding attached sensors or reflective markers to reduce visual artifacts while preserving bare-hand interaction.AprilTag markers define the planar calibration area around the hand.Source: §5.1.1 · p. 9

Appendix B.

PalmSpace was evaluated with participant-independent validation across contact, localization, and gesture tasks, then examined through indoor and outdoor ablations. The results show strong cross-user performance, with errors varying by region and design choices affecting localization and recognition.

5.1.2 Participants and Procedure.

  • The dataset contains 63,903 images from 17 participants, spanning single-finger, multi-finger, and non-contact samples.Participants performed gestures naturally without strict pose constraints; six single-finger sessions covered palm and finger regions across three modes.Source: §5.1.2 · p. 9

5.2 Validation Setup

  • LOPO validation holds out one participant per fold, testing on unseen users; localization metrics use valid detected touches and are micro-averaged across folds.Evaluation covers palm and finger touch regions, three single-finger orientations, and Scroll and Pinch gestures.Source: §5.2 · p. 9

5.3 Performance

  • 5.4 mm Palm Area MAE is lower than 7.2 mm Finger Area MAE; VT has the largest error, especially on the Finger Area.The authors attribute the region gap mainly to larger y-axis errors at points farther from the wrist-mounted camera, likely due to perspective compression; most Euclidean errors fall within 1 cm.Source: §5.3 · p. 10
  • 98.9% binary touch detection accuracy, 96.7% overall touch gesture F1, and 6.7 mm MAE summarize PalmSpace’s LOPO performance.Gesture F1 was 97.9% for Non-Touch, 97.2% for Single-Touch, 88.1% for Scroll, and 97.4% for Pinch; overall localization MAE covers valid touch samples.Source: §5.3 · pp. 9–10, Table 2 · p. 10

5.4 Ablation Study

  • The full model has the best overall indoor performance at 6.7 mm MAE and 96.7% F1, while the full architecture leads zero-shot outdoor performance among the compared architectures.Compared with unified learning, independent models increase localization MAE from 9.6 mm to 11.7 mm and reduce touch accuracy from 93.7% to 90.5%; outdoor fine-tuning reaches 6.8 mm MAE and 98.7% touch F1.Source: §5.4 · p. 11
  • Removing palm-size normalization raises overall MAE to 6.9 mm and multi-finger error from 8.3 mm to 8.8 mm, with larger errors for the three smallest-palmed users.The reported trend indicates a particular cross-user benefit for smaller palms.Source: §5.4 · p. 11, Table 3 · p. 11
  • Removing data augmentation lowers gesture F1 from 96.7% to 95.3% and raises overall MAE to 7.0 mm.The augmentations simulate camera tilt, camera–palm offset, and illumination variation.Source: §5.4 · p. 11, Table 3 · p. 11

6 User Study of Single-Finger Interaction

PalmSpace was tested for pointing, dragging, eyes-free digit entry, and outdoor use, including near-dark conditions. It approached touchscreen performance for dragging, outperformed the touchscreen on digit entry, and showed usable performance across the tested outdoor settings.

6.1 Input Methods and Participants

  • The indoor study compared PalmSpace with a smartphone touchscreen among 12 right-handed participants across four lighting settings, without individual PalmSpace calibration.Participants used the dominant index finger and the non-dominant hand for PalmSpace input.Source: §6.1 · p. 12

6.2.1 Task and Procedure.

  • The Fitts’ law task measured target acquisition and dragging with identical target parameters and external visual feedback across methods.The study analyzed 1,200 valid method-level trials, 600 per method.Source: §6.2.1 · p. 13, §6.2.2 · p. 13

6.2.2 Result.

  • 1572.2 ms total time for PalmSpace was close to the touchscreen’s 1361.6 ms, and drag time did not differ significantly between methods.Hit times were 797.7 ms for PalmSpace and 642.5 ms for the touchscreen; drag times were 774.4 ms and 719.2 ms, respectively.Source: §6.2.2 · p. 13

6.3.1 Task and Procedure.

  • The eyes-free task mapped twelve finger knuckles to a keypad layout and tested five rounds of random 10-digit entry.Participants touched designated knuckle areas to register their application-specific mapping; this did not retrain or calibrate the sensing model.Source: §6.3.1 · p. 13

6.3.2 Result.

  • 93.71% digit-entry accuracy with PalmSpace exceeded 74.86% with the touchscreen in the eyes-free task.PalmSpace stayed above 94% across identical, adjacent, and non-adjacent transitions, while touchscreen performance declined as transition distance increased.Source: §6.3.2 · p. 14

6.4 Outdoor Robustness Validation

  • In daytime testing under challenging lighting, fine-tuning the indoor-trained model reduced mean absolute error from 9.6 mm to 6.8 mm and improved touch-detection F1 to 98.7%.At night, geometric error could not be measured. A reported value was 1509.03 ms (SE = 25.49), comparable to 1572.2 ms observed indoors.Source: §6.4 · p. 14

7 Evaluation of Multi-Finger Gestures

This section evaluates PalmSpace for discrete Scroll navigation and continuous Pinch-based map manipulation, describing their interaction logic, study design, usability, design space, and limitations. It also situates these gestures within PalmSpace’s joint modeling of touch contact, absolute on-palm location, and interaction mode for cross-user inference without per-user sensing calibration.

7.1 Interaction Logic and State Management

  • The system’s FSM maintains Idle, Single-Touch, Scroll Mode, and Pinch Mode; Scroll maps gesture displacement to four directions, while Pinch maps radial and angular displacement to zoom and rotation.Displacement magnitude suppresses small unintended Scroll motions, and interaction returns to Idle upon release.Source: §7.1 · p. 14

7.2 Experimental Setup

  • 12 participants (M = 25.0, SD = 1.87; balanced gender) completed a 5-minute tutorial, and formal trials began after they reported comfort with the gestures.All participants had normal or corrected-to-normal vision.Source: §7.2 · p. 15

7.2.1 Apparatus and Tasks.

  • List Navigation yielded 20 trials per user after five repetitions; Map Manipulation used 20 Pinch trials per participant, successful when the cursor stayed within 10% scale and 10° angular margins for 1.0 s.Source: §7.2.1 · pp. 15–16

7.3 Results

  • 97.08% overall recognition accuracy across 240 List Navigation trials, with minimal, evenly distributed confusion, showed reliable directional Scroll input.Fig. 7b presents the recognition results.Source: Figure 7 · p. 15, §7.3 · p. 16
  • 7.90 s average completion time across 240 Map Manipulation trials accompanied 4.4% final scale error and 6.1° final rotation error, both within tolerance.Rotation stabilized faster than scaling (Trot = 3.59 s vs. Tscale = 4.31 s), but had a higher overshoot rate (0.96 vs. 0.55); scaling was slower but more stable.Source: §7.3 · p. 16

8.1 Usability Feedback

  • 72.5 ± 14.0 mean SUS score (median = 70.0, IQR = 16.0) corresponded to a “Good” adjective rating, with responses indicating ease of use, learnability, and confidence.One participant completed the study while mountaineering and camping in snowy conditions, suggesting potential robustness beyond controlled indoor environments.Source: §8.1 · p. 16

8.2 Design Space

  • PalmSpace’s design space combines contact state, absolute spatial localization, and interaction modality to support eyes-free, calibration-free input, including multi-finger compound controls and cross-modal transitions.Examples include scrolling, zooming, rotation, and sliding, as well as transitions between single- and multi-finger input within a task.Source: §8.2 · pp. 16–17

8.3 Limitations and Future Work

  • The dataset captures only part of real-world hand variation, and the proxy-based multi-finger model is suited to the studied scrolling and pinch tasks; future work could model richer finger articulation.The current implementation is a tethered prototype prioritizing sensing stability.Source: §8.3 · p. 17
  • A preliminary mobile deployment test reduced model size by 3.7× and accelerated inference by 2.7×, reaching nearly 30 FPS on a Redmi Turbo 4 smartphone.Future work targets a more compact, self-contained wearable and on-device deployment through compression and system-level optimization, including quantization, pruning, and adaptation to lower-power platforms while maintaining current interaction quality.Source: §8.3 · pp. 17–18

9 Conclusion

  • PalmSpace jointly represents touch contact, absolute on-palm location, and interaction mode, enabling cross-user inference without per-user sensing calibration or separate recognition and localization pipelines.The conclusion describes this unified model as turning the bare palm into a mode-aware, body-referenced interaction surface.Source: §9 · p. 18
  • User studies demonstrate absolute pointing, eyes-free digit input, and multi-finger controls including scrolling and map manipulation.The conclusion presents these results as evidence of the potential of unified bare-palm input for future head-worn interaction.Source: §9 · p. 18

A.1 Network Configuration

This appendix section details PalmSpace’s multi-task palm-input model, real-time pipeline, and data-collection procedures, alongside supplementary evaluation findings for digit entry and handwriting. In the handwriting experiment, PalmSpace achieved 92.5% average recognition accuracy on digit and uppercase-letter input, with GPT-4o used for recognition.

A.1 Network Configuration

  • A pre-trained ViT-B/16 [10] feeds a three-layer, eight-head Transformer with shared features and two-layer MLP heads for four interaction classes and position regression.The deployed model has approximately 89M parameters and jointly predicts contact, gesture class, and absolute position.Source: §A.1 · p. 21
  • 7.2 ms average inference latency per frame corresponds to theoretical throughput above 100 Hz, matching the camera pipeline refresh rate without making inference the primary bottleneck.The model was evaluated on a Mac mini with an M4 Pro chip; the reported headroom supports stable real-time interaction.Source: §A.1 · p. 21

A.2 Data Augmentation; A.3 Training Details

  • Training augmentation randomly rotates input images within [−10◦, 10◦], translates them by a maximum relative offset of 0.1 horizontally and vertically, and adjusts brightness and contrast with offset factor 0.2.Training uses an optimizer with initial learning rate and weight decay of 1 × 10−4 and batch size 160; it runs up to 100 epochs with ReduceLROnPlateau and early stopping.Source: §A.2 · pp. 21–22, §A.3 · p. 22

A.4 System Workflow

  • The wristband illuminates the palm with infrared light, and a multi-task Transformer predicts contact, gesture class, and absolute position to drive writing, scrolling, or zooming.Single-finger coordinates render points on a canvas; multi-finger coordinates map to scrolling or zooming commands.Source: §A.4 · p. 22

B.1 Apparatus and Calibration; B.2 Ground Truth Generation

  • Ground-truth collection used overhead-camera homography calibration and MediaPipe fingertip tracking, with markerless tracking avoiding artifacts in wrist-camera infrared images.AprilTags define the physical calibration plane, while the palm is approximated as planar.Source: §B.1 · p. 22, §B.2 · pp. 22–23

B.3 Collection Procedure

  • The nine-session collection covered single-finger sliding, palm-only Scroll and Pinch, and non-contact negative samples, while allowing natural multi-finger gestures and wrist flexion.Participants aligned palms in a calibration groove, with the palm center as origin; wrist flexion was encouraged within approximately ±30◦.Source: §B.3 · pp. 23–25

C.1 Indoor Settings; C.3 Handwriting Task User Study

  • Evaluation spanned four everyday indoor settings and included a handwriting task to assess continuous-input efficiency, motivated by the high accuracy demands of legible character input.The settings included a morning office, night meeting room, evening lab, and afternoon setting.Source: §C.1 · p. 25, §C.3 · p. 26

C.2 Additional Details of the Digit-Entry Study

  • PalmSpace maintained more uniform digit performance, while touchscreen accuracy dropped from identical to adjacent and non-adjacent transitions.The comparison covers digit recognition and transition accuracy; the supplied passages do not provide the table’s numeric values.Source: §C.2 · p. 25, Table 5 · p. 26, Figure 12 · p. 26

C.3 Handwriting Task User Study

  • Participants wrote randomly assigned sequences of 20 prompted characters using their preferred palm posture.The character set comprised digits 0–9 and uppercase letters A–Z.Source: §C.3.1 · p. 26
  • GPT-4o [50] was used for character recognition because of its scale and rotation robustness and structured output.The passage contrasts this choice with OCR models Tesseract and Paddle, which are described as scale- and rotation-sensitive and requiring complex tuning.Source: §C.3.1 · p. 26
  • Participants rewrote the same 20 characters on a touchscreen, yielding 480 characters across 12 participants and two conditions.Source: §C.3.1 · p. 26
  • 92.5% average recognition accuracy was achieved by PalmSpace in the handwriting experiment.The supplied passage reports the average but does not provide further recognition metrics.Source: §C.3.2 · p. 26

C.3.2 Results.

PalmSpace handwriting performance was comparable to touchscreen input, while participant feedback highlighted proprioceptive cues during eyes-free use. The supplementary sections describe pinch parameterization and outdoor robustness evaluation protocols.

C.3.2 Results.

  • 26 of 36 characters were recognized at 100%, though ‘0’ was often confused with ‘D’ or ‘O’.PalmSpace is calibration-free and supports 2D finger tracking and touch detection for handwriting, unlike systems constrained to continuous strokes.Source: §C.3.2 · pp. 26–27
  • PalmSpace handwriting performance did not differ significantly from touchscreen input (94.58% touchscreen accuracy; F1,11 = 1.21, p = 0.295).The authors characterize performance as comparable to touchscreens; palm curvature made stable strokes harder to maintain than on a flat surface.Source: §C.3.2 · p. 27

C.4 Qualitative Feedback

  • Participants reported stronger proprioceptive cues with PalmSpace than touchscreen during eyes-free use and found palm positions easier to localize than visually unavailable screen positions.They saw promise for relaxed or immersive scenarios where taking out a phone or shifting visual attention to a display is undesirable.Source: §C.4 · p. 27

D.1 Interaction Parameterization for Pinch

  • Pinch interaction represents the predicted point by radial distance from the palm center and angular position, assigning distance to scaling and angle to rotation.This decomposes one palm-based multi-finger gesture into two continuous control channels.Source: §D.1 · p. 29

D.2 Transfer Functions for Continuous Control

  • Separate transfer functions stabilize continuous interaction and improve controllability; scaling updates incrementally with radial-distance change using sensitivity coefficient λzoom = 0.3, clips scale to [0.2, 5.0], and is more stable than absolute-distance mapping.Source: §D.2 · p. 29, §D.2.1 · p. 29

D.2.1 Scaling Control.

  • The scaling and rotation controls use a rotation gain of 4.0 to support ergonomic operation with relatively small physical finger motion.The scale is clipped to [0.2, 5.0].Source: §D.2.2 · p. 29, §D.2.1 · p. 29

E Supplementary Details for Real-World Outdoor Robustness

  • The supplementary material provides outdoor evaluation methods, datasets, and quantitative ablation details for scenarios discussed in the main paper.The supplied passage does not state the ablation outcomes.Source: E Supplementary Details for Real-World… · p. 29

E.1 Daytime Data Collection and Protocol

The daytime protocol collected outdoor contact and non-contact images across varied environments and unconstrained finger postures. The accompanying ablation protocol compares indoor-trained models on outdoor data, while nighttime evaluation uses task completion time rather than direct geometric error.

E.1 Daytime Data Collection and Protocol

  • Fourteen in-the-wild daytime data sets included 10,388 contact images and 3,514 non-contact images from sunny and overcast conditions, parks, streets, and balconies.Participants touched the capturable palm area without finger-posture constraints.Source: §E.1 · pp. 29–30

E.2 Network Ablation Study

  • Zero-shot outdoor evaluation directly tested models trained indoors; the fine-tuned condition used outdoor leave-one-participant-out cross-validation.The comparison is reported with the main ablation analysis in Table 3(b).Source: §E.2 · p. 30

E.3 Note on Nighttime Ground Truth

  • Nighttime evaluation used Fitts’ Law task completion time instead of direct geometric error because RGB-based ground truth was unreliable when MediaPipe tracking degraded below 5 lux.Source: §E.3 · p. 30

F Supplementary Details for SUS Feedback

SUS feedback was favorable overall: the score corresponded to a Good usability level, with participants rating ease of use and learnability positively.

F Supplementary Details for SUS Feedback

  • 72.5 ± 14.0 was the overall SUS score (median = 70.0, IQR = 16.0), corresponding to a Good usability level.Participants rated ease of use at 5.00 and learnability at 5.17.Source: F Supplementary Details for SUS Feedback · p. 30
  • Participants rated learnability and the statement that most people would learn the product quickly at 5.17.The responses indicate favorable perceptions of ease of use, learnability, and confidence during interaction.Source: F Supplementary Details for SUS Feedback · p. 30
  • Agreement with negatively worded items remained relatively low, including 3.18 for needing technical support and 2.17 for needing to learn a lot before starting.The authors interpret these responses as indicating limited perceived friction and a relatively smooth learning curve.Source: F Supplementary Details for SUS Feedback · pp. 30–31

Related papers

Loading 2610.10370v1…