The Automated Drone Racing Heroes of the Future
For the past two years, NASA’s Jet Propulsion Laboratory in California has been dedicating its efforts towards research in autonomous drone technology. The lab’s achievements in vision-based navigation, which were funded by Google, were put to the test late this fall, in a race between one of its unmanned drones and a professional drone pilot.
The autonomous drone came up against Ken Loo in time trials this October, which were closely monitored for analysis of the behavior of both the human and artificial intelligence pilots. The space agency had developed three different drones specifically for the task, named the Nightwing, Batman and the Joker. Development was heavily focused on fine-tuning the complex algorithms needed to optimize the drones’ efficiency when navigating a high-speed, obstacle-laden race course. Special attention was given to obstacle avoidance and maximization of speed through chicanes, with help from Google and JPL’s Tango technology.
While human pilots tend to retrospectively describe their piloting decisions as ‘instinctive’ and guided by ‘the feel of things’, the artificial intelligence drones were designed to take a much more calculating approach. The differences in technique between the human and automated pilots became quite apparent during later analysis: humans tend to accelerate aggressively when given the opportunity, resulting in a somewhat jerky flight path, whereas the automated drones maintained a relatively smooth path and more stable velocity during the flight.
During the time trials, Ken Loo’s drone achieved higher maximum speeds than the automated pilot’s drone, and he was able to combine this with expert aerobatics in the hope of gaining an advantage over his machine opponent. Such exertion on Loo’s part, however, takes its toll, and it was observed that the artificial intelligence drones performed more consistently over time and never gave in to exhaustion. Speaking later, Loo admitted that exertion took its toll over one of the densest tracks he had ever flown, resulting in a loss of focus due to mental fatigue after ten or so laps. The automated pilot’s lack of susceptibility to the same fatigue ensured that it could continue competing long after Loo had lost his edge.
Race analysis showed that Ken Loo’s average speed per lap was an impressive 11.1 seconds, whereas the drones which were piloted by artificial intelligence could only manage an average of 13.9 seconds. Nonetheless, the flight paths of the automated drones were much more consistent, resulting in almost identical laps each time round. The artificial intelligence pilot’s ability to perform in a consistent manner, with immunity to mental fatigue, indicates that the drone racing champions of the future are unlikely to be humans.
The three drones developed for these time trials are capable of much higher speeds than the averages they achieved on this course. As progress is made in perfecting the flight algorithms, it is almost certainly a matter of time before human pilots can no longer out-fly their automated counterparts. When that day comes, we might look forward to an age of competition between drones developed by different companies. Instead of human names on the DRL leaderboard, we might be looking at names such as NASA and Google instead.