Mishaps are bound to happen when you’re building an entirely new kind of technology, but few founders see their prototypes fail in front of 2,500 people.
That’s what happened to Intelligent Flying Machines CEO Marc Gyongyosi during a demo at the TechCrunch Disrupt Startup Battlefield in 2016. His aerial robot, built to track inventory in warehouses, crashed for the first time on stage in the middle of his presentation.
“It was a stupid thing, too,” said Gyongyosi. “One of the electrical components we were using at the time stopped working, and it had to happen at that moment.”
The founder, who graduated from Northwestern University last month, said the crash taught him an important lesson. Although failures are sometimes inevitable, the most important thing is to not become discouraged and keep going.
Gyongyosi and his team finished the presentation and went on to make a number of business connections that are still important to his company today.
“People just don’t care whether the robot crashes or not — they care what we want to build,” he said. “We weren’t showing up with a finished product. We were at the Startup Battlefield to show what we’ve built so far, and they were impressed by that.”
IFM is building a fleet of flying robots that can collaborate to capture data in indoor environments without help from a human operator. Potential use cases include warehouse inventory tracking and construction site inspections.
From a navigation perspective, said Gyongyosi, developing robots for indoor data capture is trickier than building a self-driving car, because IFMs robots need to navigate dynamic 3D environments without the aid of external references like GPS signals.
“Indoors, things are constantly moving around and you can’t rely on any infrastructure,” he said. “Our vision is to build technology that solves this problem the way humans do, asking where it is and what it’s looking at, and making intelligent decisions based on that.”
To do so, IFM uses a suite of different machine learning techniques. Gyongyosi said one of his company’s biggest breakthroughs so far has been is the ability to teach robots new environments with limited human input.
Developing that machine learning capability is an extensive project in and of itself, but IFM builds its own hardware too. The decision to build custom hardware came early on, after Gyongyosi realized that no off-the-shelf drones could be retrofitted with the necessary equipment and still stay in the air for long enough to do the job.
IFM’s current prototype can operate for close to 30 minutes before recharging. Most off-the-shelf models start at 20-25 minutes of flight time before adding any additional equipment.
IFM’s technology has gone through an extensive evolution since Gyongyosi demoed at TechCrunch Disrupt. The software, he said, captures a lot more 3D information, and the robots have become significantly more lightweight. But the TechCrunch prototype still holds a special place in company lore.
“We still keep that robot in a box on one of our shelves with a TechCrunch sign,” he said. “It’s a reminder that we always have to work harder. We’ll probably build a showroom for all of our prototypes at some point.”
Image via Rice University.