💻 computer science

Mapping Embodied Affective Touch Strategies on a Humanoid Robot

This study demonstrates that affective touch strategies in human-robot interaction are strongly dependent on specific body regions and physical embodiment constraints, as evidenced by a user study with the iCub robot showing distinct touch dynamics and location preferences across free and constrained conditions that ultimately enhanced perceived social closeness.

Original authors: Qiaoqiao Ren, Omar Eldardeer, Francesca Cocchella, Rea Francesco, Alessandra Sciutti, Tony Belpaeme

Published 2026-05-13

📖 6 min read🧠 Deep dive

CC0 1.0

Original authors: Qiaoqiao Ren, Omar Eldardeer, Francesca Cocchella, Rea Francesco, Alessandra Sciutti, Tony Belpaeme

Original paper dedicated to the public domain under CC0 1.0 (http://creativecommons.org/publicdomain/zero/1.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to tell a robot how you feel. You have a robot that looks like a child-sized human, and you want to express emotions like happiness, sadness, love, or even just "pay attention to me" just by touching it.

This study asked a simple but deep question: Does it matter where you touch the robot, and does it matter if you are only allowed to touch one part of it?

Here is the story of what they found, explained without the jargon.

The Setup: A Robot with a "Skin"

The researchers used a robot named iCub. Think of iCub as a robot with a full-body "skin" made of sensitive sensors. It's like the robot is wearing a suit that can feel a poke, a squeeze, a rub, or a tap anywhere from its head to its toes.

They invited 30 people to come in and play a game. The game had three rounds:

The Free Round: You can touch the robot anywhere you want to show an emotion.
The Arm-Only Round: You can only touch the robot's arms.
The Torso-Only Round: You can only touch the robot's stomach/chest area.

In each round, the participants had to express eight different feelings (like "I love you," "I'm angry," or "I'm sad") by touching the robot for 10 seconds.

Discovery 1: The "Social Map" of the Robot

When people were free to touch anywhere, they didn't treat the robot like a random object. They treated it like a social being.

The "Safe Zones": People mostly touched the robot's upper body—the arms, hands, chest, and face. These are the "socially acceptable" zones, just like how we hug friends or pat a dog on the head.
The "No-Go Zones": Almost nobody touched the robot's legs, feet, or private areas (like the pelvic region). Even though it's a robot, people still respected a "social boundary."
The Surprise: Interestingly, people touched the robot's chest (torso) much more than they would touch a human's chest. With humans, we usually keep our distance there unless we are very close. But with the robot, the chest felt like a safe, central place to express feelings.

The Metaphor: Think of the robot's body like a map of a city. People mostly visit the "downtown" (arms and chest) because it's open and friendly. They rarely visit the "suburbs" (legs) or the "private backyards" (feet/pelvis), even if they are allowed to.

Discovery 2: The "Language" Changes Depending on the Body Part

This is the most fascinating part. The researchers found that how you say "I love you" changes depending on where you are saying it.

When touching the Arms: People used movement to show emotion. They rubbed, patted, or waved their hands. The motion was the message. It's like speaking with your hands; the speed and rhythm told the story.
When touching the Chest: People used pressure to show emotion. They pressed harder or softer. The force was the message. It's like a deep, steady hug where the weight of the hand matters more than the movement.

The Metaphor: Imagine the robot's arms are like a drum (you talk by hitting it in different rhythms), and the chest is like a pillow (you talk by how hard you squeeze it). You wouldn't try to tell a story by squeezing a drum, and you wouldn't try to tell a story by rhythmically patting a pillow. The body part dictates the "dialect" of the touch.

Discovery 3: You Can't Just "Copy and Paste" Touch

The researchers wondered: "If I touch the robot's arm in the 'Free Round' to show anger, will I do the exact same thing in the 'Arm-Only Round'?"

The answer was no.

Even when people were restricted to the same body part (like the arm), their behavior changed completely depending on whether they had the whole body to choose from or just the arm.

The "Free" Touch: When you have the whole robot, you might choose the arm because it feels right for that emotion.
The "Restricted" Touch: When you are forced to use only the arm, you have to change how you touch it because you can't use the rest of the body to help express the feeling.

The Metaphor: It's like writing a letter. If you have a whole notebook (Free Touch), you might write a long, flowing story. If you are forced to write on a single sticky note (Restricted Touch), you have to use different words and a different style to get the same point across. You can't just copy the long story onto the sticky note; you have to rewrite it entirely.

Discovery 4: The "Friendship" Effect

After the experiment, the researchers asked the participants how they felt about the robot.

Did they think the robot was "alive"? Not really. The participants didn't suddenly start believing the robot could feel pain or think like a human.
Did they feel closer? Yes. Even though the robot didn't talk back or hug them back (it was a one-way interaction), the participants felt closer to the robot after touching it.
The "Stranger" Factor: People who still thought of the robot as a "stranger" felt uncomfortable touching it. But those who started to see it as a "friend" felt very comfortable.

The Metaphor: Touching the robot was like a brief handshake with a stranger. It didn't make you think the stranger is your soulmate or a magical being, but it did make you feel slightly less distant from them. It broke the ice.

The Big Takeaway

This study teaches us that touching a robot is not just about pressing a button.

Location matters: You can't just put sensors everywhere and expect the same results. You need to focus on the "social" parts of the robot (arms, chest, face).
Context matters: The robot needs to understand that a "rub" on the arm means something different than a "rub" on the chest.
It builds connection: Even a simple, one-way touch can make a human feel a bit closer to a machine, provided they feel comfortable with it.

In short, to make robots that can understand our feelings through touch, we have to remember that where we touch is just as important as how we touch.

Technical Summary: Mapping Embodied Affective Touch Strategies on a Humanoid Robot

Problem Statement

Current research in Human-Robot Interaction (HRI) regarding affective touch is limited by a reliance on isolated body parts (typically forearms or hands) and predefined gestures. This approach fails to capture how affective touch generalizes across a full humanoid body or how it is shaped by the robot's embodiment, physical constraints, and perceived social role. Existing datasets often lack the spatial diversity required to understand how humans naturally distribute emotional touch across a robot's entire form. Consequently, models trained on limited data risk misinterpreting emotional intent, and tactile sensor designs may overlook regions humans intuitively use for emotional communication. There is a critical need to understand how touch strategies vary by body region, how they adapt under spatial constraints, and how one-directional affective touch influences human perceptions of the robot.

Methodology

The study employed a within-subjects experimental design involving 30 participants interacting with the iCub humanoid robot, a platform equipped with distributed capacitive tactile sensors across its arms, torso, hands, legs, and feet.

Experimental Conditions: Participants expressed eight affective states (Happiness, Anger, Sadness, Comfort, Love, Empathy, Gratitude, and "Grab attention") under three distinct conditions:
1. Free Touch: Unrestricted access to the robot's entire body.
2. Arm-Only Touch: Restricted to the robot's arms.
3. Torso-Only Touch: Restricted to the robot's torso.
Data Collection:
- Tactile Data: Continuous pressure signals (taxel readings) were captured at 8-bit resolution.
- Visual Data: Three external cameras recorded interactions, from which motion features (frame-difference energy) were extracted.
- Subjective Measures: Pre- and post-interaction questionnaires assessed perceptions of the robot's agency, communality, social role attribution, and interpersonal closeness (Inclusion of the Other in the Self, IOS).
Analysis: The study utilized mixed-effects logistic models to analyze body-part selection, eta-squared effect sizes to compare emotion-related variations in pressure and motion features across conditions, and Hotelling's $T^2$ tests to evaluate the similarity between free-touch and constrained-touch strategies.

Key Contributions and Results

1. Spatial Preferences and Body Region Dependence

The study revealed a non-uniform distribution of touch across the robot's body.

Frequency vs. Selectivity: While socially accessible upper-body regions (torso, arms, hands, face/head) were touched most frequently, less frequently touched regions (back, neck, waist) carried stronger emotion-specific selectivity. For instance, "comfort" and "love" showed strong selective associations with the back, despite low overall touch frequency.
Implication: Sensor placement should not merely maximize coverage but prioritize socially salient upper-body regions while also instrumenting selectively informative, lower-frequency regions.

2. Reconfiguration of Touch Dynamics Under Constraints

The study demonstrated that touch dynamics are not invariant; they reorganize based on the available body region.

Modality Shift: In the arm-only condition, emotion-related variation was more evident in motion features (e.g., spectral centroid, energy) than in pressure. Conversely, in the torso-only condition, pressure features (e.g., duration, gradient, intensity) were more informative for distinguishing emotions.
Lack of Transferability: Touch strategies did not transfer directly from free-touch to constrained conditions. Even when touching the same body part (e.g., the arm), the dynamic profile (pressure and motion patterns) differed significantly between the free-touch and arm-only conditions. This indicates that spatial constraints alter not just where people touch, but how they express emotion.

3. Perceptual and Social Impact

Relational Closeness: While one-directional, task-oriented touch did not significantly alter participants' perceptions of the robot's agency or capacity for feeling (Agency & Experience scales), it significantly increased perceived relational closeness (IOS scores).
Social Role Shifts: Approximately 30% of participants shifted their perceived social role of the robot toward more affiliative categories (e.g., from "stranger" to "friend" or "relative") after the interaction.
Comfort: Comfort levels were strongly correlated with social framing; participants who maintained a "stranger" classification reported significantly lower comfort than those who adopted friendlier roles.

Significance and Claims

The paper argues that affective touch is fundamentally embodied and body-region dependent. The authors claim that:

Touch strategies cannot be treated as location-invariant signals. Models for emotion recognition and generation must explicitly condition on body region and adapt their representations of pressure and motion based on the specific anatomical context.
Embodiment shapes expression. The robot's physical form, accessibility, and perceived social role actively shape the emotional meaning and physical characteristics of human touch, rather than touch being a context-free signal.
Design Implications. Tactile sensor design should move beyond sparse, local point contacts to support richer spatiotemporal interactions, balancing high-sensitivity coverage of frequently used regions with instrumentation of selectively informative, less frequent regions.
Social Impact of Unidirectional Touch. Even without reciprocal feedback, affective touch can modulate the relational distance between a human and a robot, making the robot feel socially closer, though it does not necessarily convince users of the robot's sentience or agency.

The study concludes that affective touch should be understood as an embodied, context-sensitive communication process, necessitating a shift away from fixed gesture mappings toward adaptive, region-aware models in affective computing systems.