# EmbraceMe – An Inflatable Soft Robot for Emotional Care

> Engineering a soft-robotic hug — and reporting where it fell short.

- **Case study:** https://yazdanjoo.de/projects/embraceme-soft-robotics
- **Author:** Sanaz Yazdanjoo (UX Engineer)
- **Role:** HCI Researcher & Prototyping Engineer (team of 3)
- **Year:** 2023
- **Timeline:** Summer semester 2023
- **Status:** published
- **Context:** University research project
- **Summary outcome:** The hug delighted visitors — but without signifiers nobody could start one unprompted; the interaction, not the hardware, fell short.

A Pneumatic Soft-Robotic Hugging Interface, Exhibited & Evaluated in Public

## About

A soft-robotics project exploring whether a machine can deliver the calming effect of a hug. In a team of three I designed and built EmbraceMe, a standalone hugging robot with inflatable PneuNet foam arms driven by an Arduino and a capacitive touch sensor. We exhibited it publicly and observed real visitors using it — documenting both what delighted them and where the design fell short.

## Challenge

Physical touch like hugging releases oxytocin and measurably reduces stress — but not everyone has access to it: loved ones may be distant, or contact may be unsafe. We set out to build an inflatable soft-robotic interface delivering Deep Pressure Stimulation through a standalone hugging experience, gentle enough for direct human contact.

## Solution

A standalone hugging bot built on a mannequin frame with PneuNet bending-actuator arms: foam limbs (80×16 cm) with 45° triangular cuts housing heat-sealed TPU air chambers. A 5-pad capacitive touch sensor on the chest triggers inflation when a user leans in for a hug; a Festo 5/3 solenoid valve and timed Arduino logic (3s inflate, 9s deflate) regulate pressure, with an LED feedback cycle (green: ready, white: hugging, blinking red: resetting) communicating system state.

## Methodology

We grounded the design in a literature review of interpersonal touch, Deep Pressure Stimulation, and soft robotics, and a comparative analysis of prior hugging systems (Hug Over a Distance, Huggy Pajama, HugShirt, HuggieBot 3.0, MIT's Huggable, and Bauhaus's own Hugging Suit) to identify their gaps — partial body coverage, static holds, and no emotional context. Two actuation techniques were evaluated (soft-growing vs. PneuNet bending); the standalone form factor decided for inclusivity determined the PneuNet approach. The prototype then went through iterative material testing before public exhibition with observation and user feedback.

## Process

- **discover: Deep Pressure Stimulation & Interpersonal Touch** — Reviewed research on interpersonal touch, oxytocin response, and DPS therapy (weighted blankets as the canonical example), alongside soft robotics and shape-changing interface literature.
  - Insight: Hugging is one of the most desired affectionate touches, with measurable stress-reduction effects — but replicating human hugging exactly was out of scope. The goal became a soft, warm embrace, not a simulation of a person.
- **discover: Mapping the Gaps in Prior Hugging Systems** — Analysed Hug Over a Distance, Huggy Pajama, HugShirt, HuggieBot 3.0, MIT's Huggable, and the Bauhaus Hugging Suit — comparing wearable vs. standalone forms, actuation, and sensing.
  - Insight: Every prior system shared two gaps: partial-body sensory coverage and zero emotional context. Devices sense touch, not feelings — this framed both our prototype and my later data physicalization critique.
- **define: Backpack vs. Standalone** — Two concepts sketched: a portable backpack using soft-growing (vine robot) arms, and a standalone figure using PneuNet bending actuators. Chose standalone after consultation, for inclusivity across body sizes and open access at the exhibition.
  - Insight: The form-factor decision cascaded into the actuation technique: standalone required PneuNet bending arms that hold their position in space without a skeleton — the project's hardest engineering constraint.
- **design: Balloons → Transparent TPU → Yellow TPU** — Proof-of-concept with balloons in slotted foam validated the bending mechanism. Transparent TPU replaced them for durability — but deformed irregularly after repeated inflation, breaking the curvature. Final iteration: stiffer yellow TPU chambers with regulated airflow to prevent bursting.
  - Insight: Material properties drove interaction quality: consistent chamber volume was the difference between a controlled embrace and an erratic one. Slot spacing (10 cm) and chamber size (8×6 cm) were tuned empirically across numerous arm samples.
- **design: Touch-Triggered Hugging with an LED Feedback Cycle** — A 5-pad capacitive touch sensor on the bot's chest detects a user leaning in; Arduino opens the Festo 5/3 valve for 3 seconds to inflate, holds pressure during the hug, then vents for 9 seconds on release. LED states (green/white/blinking red) communicate readiness, hugging, and reset.
  - Insight: Timings were derived through trial-and-error experimentation — long enough to hold a firm hug, short enough to protect the chambers from over-inflation.
- **deliver: What Visitors Actually Did** — Exhibited at a university event open to academic and non-academic visitors. Observed interactions and collected impressions: delight at the bot 'waking up,' but also uncanny-valley reactions to its form and confusion about how to initiate contact without our explanation.
  - Insight: Two design failures: aesthetic choices triggered avoidance in some users, and the interface lacked signifiers — curiosity did not translate into interaction without designer intervention.
- **deliver: From Binary Trigger to Emotional Data (Individual Report)** — Applied Offenhuber's data physicalization framework to critique the prototype: it followed a physical process triggered by binary signals and conveyed no emotional message. Proposed a redesign where emotion-recognition data (wearables, mood tracking) drives soft-growing arms — extension length and warmth mapped to the user's distress level, hug rhythm synced to heartbeat.
  - Insight: The reflection reframed the project's failure as a data problem, not a hardware problem — and defined ethical guardrails: user autonomy over hug intensity, explicit consent, and privacy-first handling of emotional data.

## Results

At a public university exhibition, visitors described the interaction as fun and surprising — the bot 'waking up' to hug back was the standout moment. The evaluation also surfaced honest design failures: some users read the pink, muscular arms as uncanny and avoided full contact, and without clear signifiers, nobody could guess how to initiate a hug unprompted. My individual follow-up applied Offenhuber's data physicalization framework to propose the next iteration: emotion-recognition data (wearables, mood tracking) mapped to arm extension and hug intensity, turning a binary-triggered mechanism into an emotionally adaptive interface.

## Limitations

- The exhibition evaluation was open observation and informal feedback, not a protocol: no task set, no structured interview, and no session notes were retained that would let a finding be recounted or a frequency stated.
- Nothing about the calming effect was measured. Deep Pressure Stimulation carries that claim in the literature; this prototype produced no physiological data and no self-reports of its own.
- Exhibition visitors are self-selecting and were watched in a social, playful setting — close to the opposite of the private moment of distress the concept was written for.
- There was one build and no comparison condition — no human hug, no static weighted alternative — so the observed reactions cannot be attributed to the soft-robotic actuation rather than to the novelty of the object.
- Arm length, cut geometry and the 3s/9s inflation timing were fixed for one body size. Fit, pressure and comfort across different bodies — and for the impaired users the concept names — were never tested.
- The redesign proposed in the individual reflection report — emotion-recognition data driving arm extension and hug intensity — was never built or evaluated; it remains an argument, not a result.

## Key Numbers

- **Public** — exhibition with live user observation
- **3** — prototype iterations to a working hug
- **3s / 9s** — inflate–deflate cycle timing
- **80×16** — cm PneuNet foam arms

## Outcome

EmbraceMe was not iterated into a second physical prototype after the exhibition — the group deliverable ended at the evaluated build. The follow-on work is a documented redesign proposal (individual report), not a shipped or roadmapped system, and no organisation has committed to building it.

- The visible design failures (uncanny arm aesthetics, no signifiers for initiating a hug) were not patched within the group project — they were carried into an individual critical-reflection report instead.
- The redesign proposal rejected the binary-trigger mechanism outright, replacing it with emotion-recognition data driving arm extension and hug intensity, with user autonomy, explicit consent, and privacy-first handling defined as guardrails before any such system would be built.

## Methods

- Literature Review
- Comparative Analysis of Prior Systems
- Iterative Physical Prototyping
- Material Testing
- Public Exhibition & Observation
- Critical Reflection (Data Physicalization Framework)

## Tech Stack

- Arduino
- Festo 5/3 Solenoid Valve
- 5-Pad Capacitive Touch Sensor
- TPU (heat-sealed)
- Foam Fabrication
- LED Feedback

## Skills & Topics

Soft Robotics · Human-Robot Interaction · Shape-Changing Interfaces · Literature Review · Competitive Analysis · Physical Prototyping · Material Testing · Arduino · Sensor Integration · Interaction Design · Data Physicalization · Exhibition Research