Automation has come a long way since James Watt developed the self-regulating steam engine that drove the industrial revolution 250 years ago.
A rare earth magnet motor has around 5,000 times the power density of those old engines, and gathering all the people on earth would produce just a fraction of the computational power of a microprocessors in a smartphone.
The coming machine revolution promises an altogether different locomotive to the steam machine that powered the first. Flying machines are coming to change the world.
Closing the loop
Drones that are controlled by open-loop systems can be manoeuvred manually, but lack a feedback circuit that would enable them to react and adapt to disturbances automatically.
A typical example of an open-loop system is a tumble dryer that stops at the end of a 30 minute cycle, regardless of whether the clothes inside are dry or not.
Raffaello D'Andrea is finding new ways to close this loop. The professor of dynamic systems and control at ETH Zurich is helping machines influence the physical world through advances in sensors, computation power and communication.
“We can close this loop at very, very fast rates.” D’Andrea explained at the GOTO Accelerate 2016 Business Conference. “This is what's unprecedented.”
He demonstrates the process with a small electronic cube. A combination of sensors including accelerometers and rate gyros determine the machine’s orientation like ears do for a human, while motorised reaction wheels adjust its balance and adaptation algorithms let it learn from practice.
When he pushes the cube once, it topples onto its side. But when he prods it again, the motors start to whirr and ‘Cubli’ slams the brakes. It tilts onto its tip and stays there, balancing on the edge.
The 49-year-old has been at the centre of the recent rise of the robots. In the late 90s, as a young assistant professor at Cornell University, D’Andrea became the faculty advisor and system architect of the Cornell Robot Soccer Team.
In 2003, the year Cornell won their fourth RoboCup in five years, D’Andrea met an entrepreneur called Mick Mountz. The MIT and the Harvard Business School graduate wanted to use robots to move inventory in his grocery delivery startup.
D’Andrea left his sabbatical at MIT to join Mountz in starting what later became Kiva Systems. In 2012, Amazon acquired Kiva for $775 million and renamed it Amazon Robotics.
The Amazon warehouses of today are managed by an army of robots that Kiva developed, packing delivery orders and adapting in real-time to changing demands through a process called 'distributive intelligence'.
Before the robots begin their new jobs, they learn their role by monitoring and adapting to warehouse situations.
“This is the only economically viable way to make a system of thousands of robots,” says D’Andrea. “We have to make inexpensive machines that learn how to behave like expensive machines.”
Cusp of a revolution
D’Andrea had built his first flying machine at the turn of the millennium, but put his plans on hold while working with Kiva. In 2007, D’Andrea returned to academia to take up the post of professor of dynamic systems at ETH Zurich.
In his new home in Switzerland, D’Andrea dedicated his time to pushing back the technological boundaries of drones in a large, indoor space called the Flying Machine Arena.
They’re boundaries that are expanding at a rapid rate. With trials underway in applications for agriculture, Amazon deliveries and armies, drones have been heralded as both the saviour and destroyer of humanity.
They will continue to grow regardless. The Federal Aviation Administration (FAA) expects sales of unmanned aerial vehicles (UAV) to rise from 2.5 million in 2016 to 7 million in 2020.
In his Zurich laboratory, D’Andrea and his team are shaping the future. After a ball is thrown towards a racket, a machine generates thousands of trajectories for how it will be returned. The results create a feedback strategy.
“We pick the best one, and we employ that strategy for a short amount of time — 20 milliseconds,” says D’Andrea. “Then, 20 milliseconds later, we do the whole thing again. We figure out this is what the world is like.”
The strategies are not static. They change as the world changes, and adapt to conditions. When a flying machine follows a figure-eight circuit, it uses adaption algorithms to create more accurate trajectories with every lap.
That knowledge it gains can be applied to different problems. They repeat the circuit, this time with the UVA balancing a pole on its roof. After 20 laps, it’s smoothly rounding the track at speed.
The main cost of drones, says D’Andrea, is the replacement of the batteries rather than the energy. They cost roughly one cent per kilometre per kilowatt.
A greater concern is the safety of drones, an issue that almost had catastrophic consequences during an alpine slalom race in Italy late in 2015.
"The world would be very different if that flying machine had hit that poor fellow," says D’Andrea. "Society is very reactive. Nobody talks about drone safety. It's a very important thing.
"If that had hit that guy, regulations would be all over us. It's really up to the folks that develop flying machines to be proactive about this."
His Verity Studios has designed flying lampshades for a Cirque Du Soleil show on Broadway called Paramour. They hover above the performers in front of 2,000 people eight times a week.
"It needs to work every single time," he says. "So many things on a drone can fail. Battery cable, communication can be lost."
The Paramour drones have the built-in redundancy of two batteries and two flanking cases to avoid a disaster if a malfunction occurs.
D’Andrea has developed UAVs that can stay stable and land safely even if they lose power or are hit by an object. But not everyone is taking such caution.
"People want the problem solved. Technology companies are interested in solving them, but they’re interested in the short term," he says. "Get money. Raise money. Build the team and hope that there's a market for it."
Risk mitigation is all too easily ignored. "You're going to see a lot of startup companies in this space that will not be successful for that very reason," he warns.
The game changer
Machines that learn have enormous potential, but will struggle to find a market if they need years to fulfil their potential. Put them in teams and the improvements can be exponential.
"What we can do with machines is that we can have them collaborate and communicate with that information,"says D’Andrea.
"If you had 1,000 such machines that were sharing information, it would take them less than a day to do what, on its own, it would take them a year to be able to do or more."
D'Andrea should know, because it happened at Kiva. They shared what they learned, leading the performance of the warehouse to quickly improve.
The system that controls them can be monitored and adapted as they learn, and what's discovered in one warehouse can be applied to another.
The machines are designed to get better with time, with systems controlled by embedding optimisation in a feedback loop.
"It's a very radical way of designing systems and machines." says D'Andrea. Right now, we don't think about machines in that way at all. We think of machines that degrade over time. We don't think about [them] as improving over time."
Read next: Seven drone startups to watch in 2016
He doesn't set rules to build autonomous systems, but D'Andrea does warn against one particular strategy: designing machines that imitate nature.
"Anthropomorphic robots are woefully limited because nature is playing by very different rules than technology," he says.
"Life is playing by different rules. Life is going forward. We were not designed to be what we did. We evolved to do that.
"We have very different sets of rules, technology, and we're trying to duplicate what it is that nature finds easy. We shouldn't be doing that."
D'Andrea's designs might not imitate nature, buy they do take their lead from it: adapting to surroundings, and always going forward.
Find your next job with techworld jobs