It’s Man vs Machine Again in NASA’s Drone Race

Fagjun | Published 2017-11-30 11:13

NASA has pitted professional drone pilot Ken Loo against artificial intelligence to test how a new drone racing software competes against humans.


Video by NASA/JPL-Caltech

When it comes to man vs machine, there’s no clear winner. Artificial intelligence can beat humans at some things, like the ancient Chinese game Go, but humans are the clear winners at other things. Both humans and AI have different sets of strengths and limitations, which makes competitions between the two deeply interesting.


October 12 at NASA’s Jet Propulsion Laboratory (JPL) saw another round of humans against artificial intelligence. NASA had invited drone racer Ken Loo over to test how well a new navigation software will perform against a human racer.  JPL had been making leaps and bounds when it comes to vision-based spacecraft navigation--accomplishments that attracted the attention of Google. Google was then responsible for funding the two-year-long research on making drones autonomous.


So in this drone race of man against machine, who won?

The Algorithm

"We pitted our algorithms against a human, who flies a lot more by feel," said Rob Reid, who headed the project. JPL built three custom drones--which they named Batman, Joker, and Nightwing--and developed the complex and necessary algorithms for the drones to navigate their way through the race. The drones were built for racing, and could go up to 129 kilometers per hour in a straight line. However, the obstacle course at the JPL warehouse will force the drones to fly at only 48 to 64 kilometers per hour before it’s necessary to hit the brakes.


Each drone was fitted with two cameras, which help the algorithms compare a pre-loaded map of the obstacle course to what the drones “see”. The augmented reality technology Google Tango made it possible for the drones to use their vision to determine where they are in the obstacle course.


What, therefore, is the difference between a human drone pilot and an artificial intelligence program? Loo was able to learn what the obstacle course was like after several laps. Thus, he was able to hit higher speeds, and was also able to take risks by making his drone perform remarkable maneuvers. The drones, meanwhile, were more consistent but also more cautious than Loo.

Human Limitations

There were also times when the AI-piloted drones moved so fast that all their cameras caught was a blur, thus confusing their computers. Thus, there still need to be some tweaks in the system to address this problem, since a racing drone can’t be having vision problems from flying too fast.


However, Loo experienced his own problems as well. He eventually had to battle mental fatigue, which will not be a problem for drones. “One of my faults as a pilot is I get tired easily,” Loo said. “When I get mentally fatigued, I start to get lost, even if I've flown the course 10 times.” Loo’s times were also less consistent than those of the drones.


Loo’s average time was 11.1 seconds, while the drones clocked in at 13.9 seconds. Thus, man won this round of man versus machine, but who knows what machine can dish out in the future.

Hey! Where are you going?? Subscribe!

Get weekly science updates in your inbox!