Crawling
under
my skin

Soft pneumatic arm on a wall.
Senses its environment. Responds.

the concept visualised · after one week of phase 1

Aim

Phase 1: material · printing · motion

skin and decision layer: later phases

Concept and theory

CONCEPT

sensory intelligence

senses and responds through the arm

filter + overlap

what counts, what collides

response

how the arm answers input

↑↑↑
THEORY

extended mind

Clark, Chalmers
the mind reaches into tools

soft assembly

Clark
simple local rules, behaviour emerges

perception through the body

Merleau-Ponty
the body shapes what is sensed

Body

  • octopus-like arm, thick at wall
  • pneumatic: chambers, pump, valves
  • soft body: cast or printed
  • exposed spine below the arm
  • one vertebra per chamber
  • form not final

arm base

three chambers

EXPOSED SPINE · ONE VERTEBRA PER CHAMBER
↓

valves x3

one per chamber

↓

esp32

sensing, relays, logging

↓

pump

damped feet

↓

12 v

mains enclosed

Sensing

SENSESHOW · OPTIONSBEHAVIOUR · EXAMPLES

people

proximity, direction, touch

conductive thread zones, or hidden module

someone close → arm responds

sound

loud events, voices, room

microphone: level, onsets

too loud → pulls back

light

on, off, dim

light sensor

lights off → sleep mode

All open. Also: temperature, vibration. Strip test decides thread or module.

refs: Silva-Plata 2025 (touch stays stable under deformation) · Zahir 2026 (yarn proximity 40–60 mm) · capacitive yarn on a soft robot

Sound

I mechanical

pump, valves, air

II ambient

the room, quiet

III human

voices near the wall

Drawings: level over time, not audio. Features only. Pump masked while running.

Data flow

ignore · respond
spine

sense

environment

→

read

ESP32, features

→

decide

filter, overlap, rule set

→

switch

relays, valves

→

move

pump, arm

↓

log

CSV, features only

→
↑ rules

lab

rule sets compared

Body prototypes

EXPERIMENTS · THREE CHAMBERS AT 120° EACH · NOTHING MADE YET

P1

cast

  • printed mould, cast
  • tapered, 3 chambers
from

TeMoG 2021 · OpenSCAD moulds for three-chamber tentacles

Papadakis 2024 · octopus arm, Ecoflex 00-10, PLA moulds

P2

printed TPU

  • printed in one piece
  • bellows chambers
from

Moratuwa 2025 · monolithic printed TPU, three independent chambers

Meitani & Nisar 2025 · TPU 60A / 70A, 100–300 kPa

P3

strain layer

  • wrap on P1 or P2
  • thread, fabric, fibre
from

SoPrA 2021 · fibre-reinforced chambers, tapered section

Festo OctopusGripper 2017 · Lycra around silicone limits swelling

Xie 2020 · taper angle tunes curvature

P4

bundle + fill

  • 3 printed tubes, filled
  • centre channel for air
from

SoPrA 2021 · three chambers bundled, silicone poured around

Oxford modular actuator · centre channel for pneumatic lines

COMPARE · BEND · FORCE · SPEED · CYCLES · BURST · SAG

details if asked
  • Ecoflex 00-10 to 00-50; harder silicone needs 5–6 bar vs 1–2 bar (Papadakis)
  • TPU 60A or 70A, not 85A; worked at 100–300 kPa; about 10 h per print; dry filament first (Meitani)
  • TPU gripper vs silicone gripper: 5× load, 2× bend (a gripper, not an arm)
  • resin can block silicone cure: cure test first
  • three chambers 120° apart bend in any direction (IJRR soft-arm bending)
  • no source combines taper + 3 chambers + printed TPU: the prototypes combine pieces
  • details are from abstracts and search summaries: open the papers before citing

Motion

reach

pause

retract

measure: pressure → bend angle, tip position, speed
set: hold time, distance, retract speed

reference: 10 ms pulses, 0.1–0.5 bar, up to 24% elongation (Papadakis 2024). Valves on/off, so motion is stepwise.

Previsualisation and next steps

[ render: arm and spine on a wall, visitor approaching ]

NEXT TESTS · NO RESULTS YET

01

material route

cast vs printed

02

single chamber

pressure → bend

03

one electrode

thread on a strip

04

one sound test

level, onsets

Wall height, arm length, room: undecided. Failures count as results.