Oliver Barrow

Designing physical products and their software. Founder of Jellybots. Previously Beta Firm & Barrows.

Render of the GlobalRelief unit set up for power: its four solar panels laid out either side, a laptop and phones plugged into the sockets on its front

The 3D Printing House: a lorry's container unfolds into a gantry carrying a cement print head, and the walls of a small house are being printed beneath it

Scale model of NASA's Mars Ingenuity helicopter

MOD-U-LAR, the redesigned OMO bottle and what it builds into: the labelled bottle, two bottles snapped together, a raised garden bed, a succulent planter and a block game, all made from the same blue bottles

Canopy Rush, a jungle endless runner the Game Factory made from one sentence: a monkey runs past temple ruins and waterfalls, with the score, hearts and timer of the shared game engine

Personal research · 2024 – 2026 · Experimental

Inverse Design for Inflatable Structures

Give it the shape you want and it evolves the stitch pattern for a heat-sealed inflatable that inflates into that shape.

The 2 m Pringle the tool was given, the spiral of seams it evolved for it, and the inflated film the simulation produced — Target, evolved template, inflated film: a 2 m Pringle, the spiral of seams the tool found for it in fourteen generations, and the film that results, 48 mm from the saddle on average

The inflatable is two layers of film cut flat to the same outline and sealed together around the edge and along stitch lines. The channels between the lines gather the sheet as it inflates, and that gathering is what gives it a shape. The tool flattens the target into the cut outline, lays stitch lines along the direction the sheet needs to gather, inflates every pattern in a physical simulation of 75 µm polyethylene film at 2 kPa, and scores the result by its mean distance from the target.

Patterns then evolve over generations: the best are kept and the rest replaced by mutated and crossed copies, with mutations that widen or narrow the seals where the parent's inflated film came out too far from the target. On a 7.8 m hemispherical shelter the mean error fell from 222 mm to 173 mm in 22 generations, on a 2 m four-lobed dome from 70 mm to 46 mm, and on the 2 m Pringle above from 81 mm to 48 mm in 14 generations. The dome's evolved pattern is a ring of radial gores, as on a parachute.

Each run ends with a full-size SVG template to cut and seal, the inflated shape as an STL and pictures comparing it with the target. It is written in Python with NumPy, SciPy and Numba, and the film solver was checked against the classic Mylar-balloon solution before being trusted. An earlier version used Blender's cloth simulation, and a side experiment made a two-chamber canopy walk by inflating its chambers in turn.

Input: An STL of the shape: an open, curved sheet such as a dome or canopy
Output: Full-size cut-and-seal template (SVG) and the inflated shape (STL)
Stack: Python, NumPy, SciPy, Numba; a membrane solver of my own
Status: Research prototype; resolves coarse stitches only so far