The paper asks whether artificial cilia can approach natural cilia in size, speed, 3D non-reciprocal motion, individual control, durability and compatibility with salty water. The authors report (measured) that acrylic acid-co-acrylamide hydrogel microcilia (2–10 µm wide, 18–90 µm tall), 3D-printed by two-photon polymerization with nanometre-scale pores, bend in response to 1.5 V without electrolysis because cations migrate inside the gel: H+ accumulation shrinks the cathode-facing side in deionized water (bending toward the cathode), whereas Na+ drags water and swells it in physiological sali

Live model of a 5 × 5 array · each hair spins ~20 times a second in the experiments; shown slowed · flow is a simplified illustration

Between columns 2 and 3: water flows ↑ up · centre: ↻ clockwise swirl

Nature · January 2026 · Liu, Wang, Ren … Hu & Sitti

Your airways and a growing embryo are lined with cilia — microscopic hairs that row fluid in coordinated patterns. This team 3D-printed artificial ones from soft hydrogel, each wired to its own four electrodes. Tap a hair to reverse its spin and watch the water re-route.

Program

↻ clockwise · ↺ anticlockwise · · off Row 1 on top, column 1 on the left.

Rowing through honey

Cilia are everywhere in biology. They sweep mucus out of your airways, move cells along the fallopian tubes, help neurons mature and, in the embryo, set up the flow that tells left from right. Starfish larvae use them to swim and to feed.

At this size water behaves like thick honey. In the paper's setup the Reynolds number — inertia compared with stickiness — is about 0.001: nothing coasts, and everything stops the instant the push stops. So natural cilia beat with non-reciprocal strokes: the way back is different from the way out.

Honey rule: here, a stroke that exactly retraces itself undoes itself.

Beat 0 · watch the amber tracers and the rings where each beat started.

So an artificial cilium must be tiny, soft, quick — and able to sweep in 3D. Why has that been so hard?

The wish list nobody could tick

Engineers have built artificial cilia driven by air pressure, magnets, ultrasound, light, static electricity and electrochemical reactions. Each ticks some boxes on the wish list below and misses others. Pick an approach and see which requirements light up.

This work: all seven, at 1.5 volts.

Based on the paper's comparison table (Extended Data Table 1). “Not shown” means the property was not reported for that approach.

Enter the hydrogel

A hydrogel is a soft, water-filled polymer network — think jelly. This one, acrylic acid-co-acrylamide (“AAc-co-AAm”), carries fixed negative charges (–COO−) along its chains. Ions moving in or out make it swell or shrink, and it is about as soft as body tissue (modulus ≈ 1,000 Pa).

But electrically driven gels have been slow — tens of seconds to minutes per bend — and have needed tens of volts: enough to split water into bubbles (electrolysis), which damages things.

The surprise in this paper: make the gel small enough and print it the right way, and both problems vanish.

Shrink it, and the ions arrive on time

A gel bends when ions pile up on one side. How long that takes depends on how far the ions must travel, and on how hard the electric field pushes them.

Field strength is voltage divided by distance. Put the electrodes close together and a battery's 1.5 volts gives a strong field; spread them out and you need hundreds of volts — far beyond the point where water splits.

10 µm: H⁺ crosses in 2.8 ms · Na⁺ in 16.7 ms

Crossing time = distance ÷ (ion mobility × field). Times scaled from the paper's values at 10,000 V/m.

1.5 V across 150 µm gives 10,000 V/m

No electrolysis

Printed with light, full of tiny channels

The hairs are written by two-photon polymerization : a laser focused so tightly that only its focal spot hardens the liquid ink, tracing the gel layer by layer (300 nm slices, 200 nm hatching, about 15 mW). The result is a network with nanometre-scale pores.

Each pore wall is charged and holds a thin cloud of counter-ions — the “electric double layer”. In nanometre pores these clouds fill much of the channel, so the field moves ions, and the water with them, far more effectively.

Qualitative sketch; driven by the same Play/Pause as the lab above.

Supplementary Video 6, Liu et al., Nature (2026), CC BY 4.0; trimmed and re-encoded.

Why opposite? Because in the tiny gel the motion comes from ions moving inside it.

Which way will it bend?

Here is one hair, 90 µm tall and 10 µm wide, between two electrodes. Negative charges are fixed to the gel's chains; positive ions float free. Switch on 1.5 V and the free positive ions drift toward the − electrode.

What happens next depends on which ions are in the water. Pick a solution, predict, then apply the voltage.

Tip direction while the field is on (0–0.8 s)

Paper (Fig. 2a): toward − at 0.032, 0.092 and 0.128 s.

Pure water: when the voltage comes on, which way will the hair bend?

Make a prediction — the result appears when the field switches off.

H⁺ ions pile up on the − side. They turn fixed –COO⁻ into neutral –COOH, so the chains stop repelling each other and that side shrinks → the hair curls toward − (cathode).

Supplementary Video 8, Liu et al., Nature (2026), CC BY 4.0; trimmed and re-encoded.

H⁺ is about seven times more mobile than Na⁺ (3.62 × 10⁻⁷ versus 5.19 × 10⁻⁸ m² s⁻¹ V⁻¹). That race explains the pinch-of-salt reversal: the fast ions shrink region 1 first, the slow ones arrive and swell it.

In the paper, the measured step responses take 0.1–0.7 s, and the authors' own computer model reproduces all three behaviours.

In centimetre- and millimetre-sized gels, the voltage splits water at the electrodes, making acid near + and base near −. Those pH gradients — and differences in osmotic pressure at the gel's surface — drive the bend: toward + in pure water in about 120 s, toward − in saline in about 30 s.

A microcilium sits in a large bath, so ions from the surrounding water quickly erase those surface differences. What remains is ions moving inside the gel. The authors also built a coupled electro-chemo-mechanical computer model — ion transport, osmotic forces and gel elasticity together — that reproduces the directions and the cathode-then-anode reversal.

Four electrodes make it spin

Give each hair four electrodes — north, east, south and west. In pure water the hair leans toward a − pad and away from a + pad. Step the polarity around the four pads a quarter-turn at a time and the lean rotates: the tip sweeps a circle, the non-reciprocal stroke that rows water.

Spin

Or set pads by hand (+ → − → off)

10 µm hair at 10 Hz: bends ≈ 112° per beat (measured)

Spin shown 20× slower. Hairs drawn at the same height for comparison. In saline the same hair would lean toward + instead (see the real video below).

Each beat, ions must redistribute across the hair. In a 2-µm hair H⁺ needs about 0.6 ms and Na⁺ about 3.4 ms; in a 10-µm hair about 2.8 ms and 16.7 ms. At 50 Hz a beat lasts only 20 ms, so the thick hair runs out of time — and the thin one bends nearly five times more.

The water also pushes back harder as the frequency rises: the “sperm number”, a ratio of drag to stiffness, grows with frequency (Extended Data Fig. 7b).

Supplementary Video 5, Liu et al., Nature (2026), CC BY 4.0; trimmed and re-encoded.

Program the carpet

Back to the 25-hair array, each hair with its own four electrodes. Every spinning hair stirs a little whirlpool around itself. So what happens where two whirlpools meet?

Columns 1, 3 and 5 spin clockwise; 2 and 4 anticlockwise. Between columns 1 and 2, will the water go ↑ up or ↓ down?

Answer to load that program and drop a red tracked particle in the gap.

Paper experiments & patterns

Between columns 2 and 3: almost still · centre: ↺ anticlockwise swirl

Simplified flow model: each spinning hair stirs a local swirl and the swirls add up. It reproduces the paper's flow patterns, not their speeds. Tap a hair to change it; tap open water to drop a red particle (up to three).

Test the rule

✓Send water ↓ down the gap between columns 2 and 3.Column 2 clockwise, column 3 anticlockwise: both push down in the gap between them.

✓Switch the centre hair off — and still make the centre swirl clockwise.Anticlockwise neighbours all around the idle centre: their pushes chain into a clockwise loop.

✓Make water run clockwise around the top-left 3 × 3 block — down its right edge, left along its bottom.Block clockwise, everything around it anticlockwise: opposite neighbours line the block's edge with a conveyor.

Paper, Fig. 4d–g: (1) programmed motion, (2) flow simulation, (3) measured PIV flow field. Cropped from Liu et al. 2026, CC BY 4.0.

Supplementary Video 29, Liu et al., Nature (2026), CC BY 4.0; trimmed and re-encoded.

Supplementary Video 30, Liu et al., Nature (2026), CC BY 4.0; trimmed and re-encoded.

What they built and measured

Durability

Still bending about 50° after 330,000 beats (20 Hz for 5 hours) — 70% of the start. No chemical reactions are involved, so the gel doesn't wear out like redox actuators; electrodes peeling off the substrate limited device lifetime.

Body fluids

At 5 Hz: pure water ≈ 147° · saliva ≈ 127° · saline ≈ 60° · DPBS ≈ 30° · serum ≈ 30° · mouse plasma ≈ 35°

It works in all of them, but weaker as the mix of ions gets more complex: different ions arrive at different times and partly cancel each other's swelling.

A counter-intuitive recipe

More charged groups → more ions → more bending? Raise the acrylic acid content and see.

15 wt%: stiffness ≈ 1,100 Pa · tip moved ≈ 50 µm (measured) · ≈ 21.5 µm in the authors' simulation

The answer is no: the gel stiffens faster than the extra charge helps.

A bendable chip

Panels cropped from Liu et al., Nature (2026), Fig. 1b, CC BY 4.0.

From 25 to a million

Supplementary Video 15, Liu et al., Nature (2026), CC BY 4.0; trimmed and re-encoded.

Supplementary Video 16, Liu et al., Nature (2026), CC BY 4.0; trimmed and re-encoded.

An artificial starfish larva

A 3D-printed larva body carries hydrogel cilia on its curved surface and reproduces the arrays of vortices real larvae use to feed — “a robotic platform for quantitative study”, in the authors' words.

Supplementary Video 21, Liu et al., Nature (2026), CC BY 4.0; trimmed and re-encoded.

Back to the carpet

You now know why the carpet works. Follow the chain — tap any link to revisit it.

What it brings closer

Measuring how much ciliary motion matters in biology — such as the feeding vortices of starfish larvae or the embryo's left–right flow — with an artificial version that can be controlled hair by hair.

New microscale devices: pumps and mixers that work in body-like fluids, micromachines (the team already drove tiny rotary and flapping mechanisms) and microrobots.

Next steps the authors name

Better performance in complex fluids, for example with a different monomer (AMPS).

Hairs 2 µm wide or thinner, so ions redistribute even faster.

3D or closer-spaced electrodes — 5 µm spacing between cells already showed negligible interference.

Sources & credits

Zemin Liu, Che Wang, Ziyu Ren, Chunxiang Wang, Wenkang Wang, Jongkuk Ko, Shanyuan Song, Chong Hong, Xi Chen, Hongguang Wang, Wenqi Hu & Metin Sitti. “3D-printed low-voltage-driven ciliary hydrogel microactuators.” Nature 649, 885–893 (2026). https://doi.org/10.1038/s41586-025-09944-6 · Read the article at nature.com

Figures and Supplementary Videos 5, 6, 8, 15, 16, 21, 29 and 30 are from Liu et al., Nature 649, 885–893 (2026), licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Changes: videos trimmed, resized and re-encoded; figure panels cropped.

Interactive models on this page are simplified teaching illustrations based on the paper's descriptions and data; data values marked ≈ were read from the published charts.