AeroVelo's optimizer designed Atlas to take the 33-year human-powered helicopter prize

On June 13, 2013, Atlas — an enormous, ridiculous-looking pedal-powered contraption — wafted into midair and made history, winning the Sikorsky Human Powered Helicopter Competition that had defeated scores of designers for 33 years. The task: hover three metres up for at least a minute without drifting outside a 10-metre square. 'Until a few weeks ago, the people who literally wrote the book on the helicopter said that this problem is too hard — that this is actually physically impossible,' said chief structural engineer Cameron Robertson.

AeroVelo's edge was choosing the right box. Other teams reasoned: 'We're on a campus, so let's find the biggest gym to build and test this aircraft in' — making the gym's size a design constraint. That constraint was artificial, so they removed it and tested in a huge indoor soccer stadium near Toronto. Freed of the gym, the design leaned on sheer size: four rotors 10 metres in diameter each — 'Atlas is the size of every other human-powered helicopter combined,' Robertson said. The team was 'not afraid to construct a Rube Goldberg machine' — an aircraft built to do one thing: win the prize.

The rest was systematic modeling. Robertson, chief aerodynamicist Todd Reichert and the team spent months digitally modeling 'structure, aerodynamics, engine thermodynamics, pilot physiology, kinematics, all these different design considerations', then programmed an optimizer to iterate on possible designs with a 'good enough' set of tolerances rather than perfect fidelity. 'Then you hit go, and it designs the helicopter,' Robertson said. 'It only takes five minutes to run on a laptop.'

Remove artificial constraints — test in a soccer stadium, not the biggest gym you have — model the physics, and let a software optimizer design the helicopter itself

It removed an artificial constraint everyone else had accepted: the test venue's size, not physics, was capping rival designs.

The optimizer turned design from intuition into search — five minutes of computation produced what 33 years of expert iteration had not.

The team's non-specialism helped: self-described non-specialists questioned assumptions helicopter designers took as given.

Winning required embracing an impractical machine — four giant rotors, useless for anything but the prize.

'Impossible' often means someone smuggled an artificial constraint into the spec: find the real constraints, model them, and let systematic search — not intuition — design the answer.

Robertson framed the method as transferable to slower-moving fields: 'We don't have 50 years to reevaluate our infrastructure, address climate change, or deal with other impossible problems. Getting 2% improvement is no longer acceptable as the right answer.' The team argued that anyone can become a better problem solver by inhabiting the same state of mind — questioning the status quo and removing artificial constraints.