You already have an account? Log In

It Takes Two Neurons to Ride a Bicycle: The Surprising Simplicity of Balance

Posted on 30.05.2026

Ask anyone who has ever wobbled down a suburban street with a parent jogging behind them: learning to ride a bike feels like a small miracle. One moment you're a heap of elbows and tears on the nature strip; the next, something clicks, and you're gliding. The strange part is that nobody can really explain what changed. You didn't read a manual. You didn't memorise equations for gyroscopic precession. Your body just... figured it out.

A recent computational experiment suggests the answer might be more humbling than we'd like to admit. According to research highlighted by Live Science, a neural network with just two artificial neurons can learn to balance and steer a simulated bicycle. Two. Fewer neurons than a nematode worm devotes to twitching. And yet that minimal circuit is enough to master a task that defeats most toddlers for weeks.

For Australians — a nation that, according to BikeRadar, has embraced cycling as one of the most science-backed activities for brain and body health — this tiny result has big implications for how we think about motor learning, balance, and the surprisingly elegant physics of the bike itself.

The two-neuron cyclist

The study reported by Live Science involved building a stripped-down artificial neural network and giving it the job of keeping a simulated bicycle upright while steering it toward a target. The researchers kept trimming neurons until they hit the floor: two. Below that, the system collapsed. At two, it could still learn — through trial and error — to lean, counter-steer, and stay vertical.

The result is striking because deep-learning systems that play chess or recognise faces use millions of artificial neurons. Balancing a bike, an activity we associate with childhood difficulty, turns out to be computationally lightweight once you understand what's really being computed. The bicycle is doing most of the work. The neurons just have to nudge.

Why bikes (almost) ride themselves

That's not a metaphor. As Canadian Cycling Magazine and TwistedSifter both reported, researchers have studied the paths traced by hundreds of unmanned bicycles — riderless bikes pushed and left to roll until they fall. In one visualisation, the trajectories of around 800 unmanned bicycles were plotted, showing how each one wobbles, corrects, curves and finally topples.

What the data reveals is that a moving bicycle is intrinsically self-stabilising over a certain speed range. The geometry of the frame, the trail of the front fork, the angular momentum of the spinning wheels — all conspire to make the bike steer into a fall, which pushes it back upright. A riderless bike, pushed at a brisk pace, will weave along for surprising distances before it gives up and falls over.

This matters for neuroscience, because it changes the question. Your brain isn't solving the full physics problem of an inverted pendulum on two wheels. The bike is solving most of it. Your nervous system just needs to provide small, well-timed corrections — which is exactly what a two-neuron controller can do.

What your nervous system is actually doing

Real biological riding obviously involves more than two neurons. The cerebellum, basal ganglia, vestibular system, proprioceptors in your ankles, and the visual cortex all chip in. But the core control loop — sense lean, adjust steering — is shockingly minimal. It's the kind of computation that small invertebrate ganglia perform every day.

This explains a few things about learning to ride:

It can't be taught verbally. The control loop runs faster than conscious thought. Telling a child to "lean left to go left" is almost counterproductive, because at speed, riders actually counter-steer — briefly turning the bars the opposite way to initiate a lean. Nobody learns this by being told; the cerebellum learns it through repetition.
It's stored in motor memory for life. Once the circuit is trained, it persists for decades without practice. The phrase "like riding a bike" exists because the underlying network is so small and so well-consolidated that it's nearly impossible to lose.
The hard part is the failure cases. Two neurons can balance a bike, but they can't dodge a parked car, anticipate a child running onto a path, or judge a wet tram track. That's where the rest of the brain earns its keep.

Why this is good news for your brain

Here's where the minimalism of the control loop becomes interesting for everyday cyclists. Because balance is largely automated, the rest of your brain is freed up to do what BikeRadar's round-up of 27 science-backed benefits describes: improving cardiovascular fitness, lowering stress, building grey matter, sharpening memory, and even reducing the risk of dementia.

If riding a bike required the kind of full-cortex effort that, say, learning a new language does, you couldn't enjoy any of those benefits — you'd be too busy concentrating. Instead, the two-neuron-shaped core of the task offloads to deep, ancient circuitry, leaving your prefrontal cortex to wander, problem-solve or simply enjoy the view of the Yarra, the Brisbane River or your local bike path.

Cycling, in other words, is the rare physical activity that is simultaneously demanding for the body and restful for the conscious mind. The neuroscience explains why so many people report having their best ideas mid-ride.

Lessons from a minimal brain

The two-neuron result is part of a broader trend in neuroscience and robotics sometimes called embodied cognition: the idea that intelligence isn't just in the brain, it's distributed across the body and the environment. A bicycle's geometry is, in a real sense, doing cognition. So is the spinning gyroscope of its wheels. The rider's nervous system completes the loop, but it doesn't have to carry the whole load.

This has practical consequences:

For robotics: Engineers designing balancing robots increasingly exploit passive mechanical stability rather than brute-forcing balance with computation. A well-designed body needs a smaller brain.
For rehabilitation: Stroke and Parkinson's patients often retain the ability to cycle even when walking becomes difficult. The minimal, deeply embedded control loop survives damage that disrupts more cortically demanding movements. Stationary-bike therapy is now a standard tool in neurological rehab.
For learning: If you're teaching a child to ride, the lesson from both the two-neuron model and the unmanned-bike experiments is the same: get the bike moving. Stability is a property of motion, not stillness. Balance bikes — pedal-free designs that let kids scoot — work because they let the cerebellum learn the actual control loop without the distraction of pedalling.

The humility of two neurons

There's something quietly humbling about the idea that a task you struggled to learn as a six-year-old can be reduced, in principle, to two artificial neurons and a well-designed frame. It suggests that what we experience as the difficulty of learning isn't the complexity of the computation — it's the slow process of tuning a very small circuit to a very specific physical system.

The unmanned-bicycle studies show the machine knows how to stay up. The two-neuron experiment shows the controller doesn't need to be smart. What you're really doing, in those wobbling first laps of the driveway, is letting your cerebellum negotiate a quiet handshake with the geometry of the bike. Once that handshake is made, it lasts a lifetime — and, as decades of cycling research confirm, it pays dividends for your brain long after you've stopped thinking about how to balance at all.

So the next time you swing a leg over a bike, spare a thought for those two neurons doing the heavy lifting. They are, in their own minimal way, what makes the whole beautiful, brain-boosting business of cycling possible.