Tactile emoticons: Conveying social emotions and intentions with manual and robotic tactile feedback during social media communications

Social touch is central to human emotional communication, yet typical digital communications (e.g., social media) primarily rely on vision and audition and lack tactile signals. Affective touch, particularly slow stroking at CT-optimal velocities (approximately 1–10 cm/s), is associated with tactile pleasure and prosocial communication and can regulate physiological and affective states. Existing mediated touch technologies demonstrate benefits but often lack neurobiological grounding and have been evaluated only at prototype and feasibility levels. Emoticon-based communication research has focused on visual emojis, which can strengthen messages in the absence of face-to-face cues, with usage patterns aligning with facial behaviors and context-dependent meanings. The present work asks whether tactile emoticons—implemented as CT-optimal stroking—can convey greater social approval and support (social intent) than visual emoticons, and whether combining tactile with visual emoticons enhances perceived social intent and modulates physiology compared to visual-only feedback. The study comprises two experiments: Experiment 1 compares manual tactile feedback (CT-optimal vs CT-suboptimal brush stroking) against visual emoticons; Experiment 2 examines whether visuotactile feedback delivered via a soft robotic sleeve (S-CAT) at CT-optimal speed enhances perceived social intent and affects heart rate, skin conductance rate, and heart rate variability relative to visual feedback alone. The authors also explore moderating roles of post valence, identification/relevance of posts, safety/trust in the device, and individual differences in touch attitudes and affective traits.

Prior work shows that touch communicates distinct emotions (e.g., sympathy, love, gratitude) and can be decoded even from strangers. CT-afferent activation via slow stroking correlates with tactile pleasantness and prosocial effects, is used spontaneously in caregiving and adult relationships, and can modulate neurobiological systems (e.g., μ-opioid and oxytocin). Mediated touch devices (e.g., Kissenger, Huggy Pajama) can increase perceived proximity and positive affect, but their implementations often lack neurobiological frameworks and remain early-stage. Visual emoticons support socio-emotional communication, are context-dependent, and tend to be used more in positive, humorous interactions, yet their efficacy compared to tactile emoticons has not been directly tested. Theoretical and empirical reviews in affective computing and haptics suggest potential for tactile integration into digital communication, but methodological rigor and CT-informed design are needed. The present work addresses this gap by leveraging CT-optimal touch properties to design and assess tactile emoticons in social media-like contexts.

Two experiments were conducted. Experiment 1: Within-subjects 2 (valence: positive vs negative) × 2 (feedback mode: tactile vs visual) × 2 (feedback support level: high vs low) design. Participants: N = 36 (after exclusions), recruited from UCL; ethics approved; £10 compensation. Stimuli: 64 Twitter-like posts (brief sentences, 32 positive and 32 negative; individual vs social contexts balanced). Visual feedback: high support emoticons (red heart or thumbs up) vs low support emoticon (neutral blue heart). Tactile feedback: manual brush stroking on the left forearm delivered by a trained confederate at CT-optimal velocities (3 cm/s or 6 cm/s; high support) vs CT-suboptimal (0.3 cm/s; low support). Touch applied within two marked 9×4 cm rectangles for 3 seconds per trial, alternating areas to minimize fatigue; pressure controlled. Procedure: Participants viewed each post; after 10 seconds the confederate delivered either visual or tactile feedback for 3 seconds; participants rated perceived 'social approval and support' (social intent) on a 0–100 scale after each trial. Deception check confirmed the confederate’s role wasn’t detected. Analysis: Multilevel modeling (R), with participant ID as random effect; main effects of feedback mode, support level, valence, and their interactions tested. Experiment 2 (preregistered: OSF https://osf.io/f9sjv): Within-subjects 2 (feedback mode: visual vs visuotactile) × 2 (valence: positive vs negative) design. Participants: N = 52 (after exclusions); additional exclusions for physiological data (HR N = 47; SCR N = 44); ethics and procedures as in Experiment 1. Stimuli: 48 Facebook-like posts (24 positive, 24 negative), selected from Experiment 1 posts with highest social intent; varied emoticon intensity (low/medium/high) and likes range (low 20–25; medium 40–50; high 80–100) for ecological validity (not modeled as covariates). Feedback: Visual-only (emoticons) vs visuotactile (same visual emoticons plus tactile feedback via S-CAT sleeve delivering CT-optimal caress-like strokes at 6 cm/s). S-CAT: silicone pneumatic soft haptic device with eight air cells producing a rippling, caress-like tracing effect; actuator area ~9×4 cm; temperature maintained at 36°C; validated to elicit higher pleasantness at CT-optimal velocities compared to fast touch. Procedure: Participants created a profile on FaceJournal, viewed each post for 6 seconds; emoticon appears and likes ramp up for 2 seconds; feedback displayed for 12 seconds. For visuotactile trials, S-CAT touch coincided with visual feedback (9th–20th second). Participants rated social intent (approval/validation for positive posts; support/sympathy for negative posts). Additional measures: identification with and relevance to posts; feelings of safety and trust while wearing S-CAT; physiological measures (HR, SCR, HRV). Analysis: MLMs with participant ID and block as random effects; tested main effects and interactions of feedback mode and valence; PCA used to orthogonalize identification and relevance due to high correlation; secondary analyses assessed moderation by rejection sensitivity and trait anxiety on physiology. Baseline corrections applied for HR and SCR; HRV computed from inter-beat intervals.

Experiment 1: - Feedback mode: Tactile feedback yielded significantly higher perceived social intent than visual feedback (β = 12.48, SE = 1.00, p < 0.001; ICC = 0.29). - Support level: High vs low support produced higher social intent (β = 3.41, SE = 1.00, p = 0.001). - Valence: Positive posts elicited higher social intent than negative (β = 7.22, SE = 1.00, p < 0.001). - Interactions: Feedback mode × support level not significant; feedback mode × valence showed a non-significant trend (β = −3.45, SE = 1.99, p = 0.083), but tactile > visual remained significant within both positive (β = 10.85, SE = 1.45, p < 0.001) and negative posts (β = 14.11, SE = 1.30, p < 0.001). Three-way interaction not significant. Experiment 2: - Overall: Visuotactile feedback increased perceived social intent over visual-only (β = 4.20, SE = 1.32, p = 0.001). - Feedback mode × valence interaction significant (β = 6.27, SE = 2.61, p = 0.016). - Planned contrasts: Visuotactile > visual on positive posts (β = 7.34, SE = 1.50, p < 0.001); no difference on negative posts (β = 1.06, SE = 1.76, p = 0.545). - Identification and relevance: Identification predicted higher social intent (β = 4.87, SE = 1.51, p = 0.001); relevance showed a trend (β = 2.51, SE = 1.47, p = 0.088). No significant interactions with feedback mode. Identification did not differ by valence. - Safety and trust (visuotactile trials): Safety showed a trend overall (β = 0.22, SE = 0.12, p = 0.061); significant for negative posts (β = 0.29, SE = 0.12, p = 0.019), not positive (p = 0.205). Trust predicted higher social intent overall (β = 0.22, SE = 0.09, p = 0.019); significant for positive posts (β = 0.25, SE = 0.10, p = 0.015), trend for negative posts (β = 0.18, SE = 0.10, p = 0.078). - Physiology: No overall differences between visuotactile and visual feedback for HR (p = 0.638), SCR (p = 0.365), or HRV (p = 0.442). Moderation: Visuotactile feedback associated with greater SCR downregulation in individuals high in rejection sensitivity (β = 0.01, SE = 0.01, p = 0.034) and greater HRV modulation in those with higher trait anxiety (across valences β = 0.0005, SE = 0.0002, p = 0.001; negative valence β = 0.0005, SE = 0.0002, p = 0.037).

Across two experiments, tactile emoticons—implemented as CT-informed affective touch—enhanced perceived social intent compared to visual emoticons alone. This effect was robust whether touch was delivered manually by a confederate (brush stroking) or mediated via a soft robotic sleeve (S-CAT), suggesting potential for integrating tactile feedback into digital communication to enrich socio-emotional signaling. The enhancement was particularly evident for positively valenced posts, whereas negative posts did not show additional benefits from visuotactile feedback in Experiment 2. This valence-specific effect may reflect differences between regulating one’s emotional state versus perceiving the sender’s social intentions, and the contextual interpretation of emojis and touch in negative scenarios. The findings align with prior evidence that CT-optimal touch conveys prosocial intentions and supports affect regulation, and extend it to the domain of social media-like interactions. The study also highlights the importance of contextual and individual factors: greater identification with post content increased perceived social intent; feelings of safety and trust in the wearable device boosted visuotactile effects, especially for negative content; and individual differences (rejection sensitivity, trait anxiety) moderated physiological responses, hinting at targeted benefits for certain users. While overall physiological regulation was not observed, the moderation results suggest avenues for personalized haptic communication interventions. The results underscore the feasibility and promise of tactile emoticons in enhancing non-verbal communication online, but also the need for careful consideration of ethical, safety, and comfort dimensions in mediated touch technologies, particularly outside controlled lab environments. Precise touch parameters (e.g., velocity, temperature) matter for perceived intent, indicating that neurobiologically informed design should guide future haptic communication tools.

The study demonstrates that tactile and visuotactile emoticons, designed with CT-optimal touch principles, can meaningfully enhance perceived prosocial intent during social media-like communications compared to visual emoticons alone, particularly for positively valenced content. Manual and robotic touch delivered at appropriate velocities conveyed clearer social approval and support. These findings contribute to the growing evidence for the benefits of affective touch and suggest that roboaffective devices could augment digital interactions. Future research should: (1) test tactile emoticons in real-world platforms with user-generated content; (2) directly compare manual vs mediated touch modalities; (3) refine touch parameters (velocity, temperature, vibration) and context-specific mappings, especially for negative scenarios; (4) examine longer mood induction paradigms and more sensitive physiological protocols; (5) explore personalization based on traits (e.g., rejection sensitivity, anxiety); and (6) address ethical, privacy, safety, and consent considerations for remote touch in social media.

Key limitations include: (1) Potential ambiguity of certain visual emoticons in negative contexts in Experiment 1 (e.g., red heart to a negative post), addressed in Experiment 2 but still highlighting context sensitivity; (2) Inability to statistically compare social intent levels across experiments due to different samples and designs, precluding conclusions about preferences for manual vs robotic touch; (3) Use of standardized posts rather than participant-authored content may limit ecological validity and personal relevance; (4) Physiological measures may have lacked sensitivity given brief stimuli and technical issues reducing sample sizes; (5) No direct comparison between manual brush stroking and S-CAT-mediated touch within a single design; (6) Mediated touch outside lab settings raises ethical concerns (safety, unwanted touch, consent) that were controlled here but need thorough exploration for real-world applications.