00:00 Hi, you are watching InDepth with Elo.
00:09 In a stunning development, an AI drone pilot has decimated the best of human drone pilots
00:15 in a competition.
00:17 Three world champion drone pilots were recently defeated in a competition by an autonomous,
00:22 artificial intelligence-powered drone.
00:25 This is the first time that a drone powered by AI has beaten a human champion in a real
00:31 physical sport designed for, and by, humans.
00:34 The AI-equipped drone, developed by researchers at the University of Zurich, came out on top
00:41 in 15 out of 25 races and recorded the single fastest lap time at 17.47 seconds.
00:49 That brisk lap time was nearly half a second better than the best of human drone pilots.
00:55 The three human competitors, Alex Vanover, Thomas Bitmata, and Marvin Schieper, have
01:01 each won drone racing championships in the past.
01:04 All three expert drone racers were beaten by an algorithm that learned to fly a drone
01:10 around a 3D race course at breakneck speeds without crashing.
01:15 The name of this artificial intelligence-powered drone is Swift, and the competition in which
01:20 it prevailed was the FPV drone racing.
01:24 Swift has been created jointly by the researchers at the University of Zurich and Intel.
01:30 FPV drone racing is a thrilling sport where competitors control high-speed drones through
01:36 complex obstacle courses.
01:39 Pilots, wearing headsets, remotely control the drones while viewing a live video feed
01:44 from the drone's onboard camera, offering an immersive first-person perspective.
01:50 This race involves humans piloting small quadcopters around a course with a speed of over 100 km/h
01:57 with the vehicles being subjected to g-forces of up to 5G.
02:01 Swift Drone combines machine vision with real-time data capture through an onboard camera, mirroring
02:07 the setups human racers use.
02:10 Adding to its suite of features, Swift boasts an integrated inertial measurement unit gauging
02:15 the drone's speed and acceleration.
02:18 All this information feeds into an artificial neural network that computes the drone's location
02:24 in real-time and discerns race gates.
02:27 Furthermore, a deep neural network-based control unit determines the optimal racing path, ensuring
02:34 Swift completes circuits rapidly.
02:37 Swift used a technique called deep reinforcement learning to find the optimal commands to hurdle
02:43 around the circuit.
02:44 Because the method relies on trial and error, the drone crashed hundreds of times in training.
02:51 But since it was a simulation, the researchers could simply restart the process.
02:55 Analysis of the races showed that Swift was consistently faster at the start of a race
03:01 and pulled tighter turns than the human pilots.
03:04 The quickest lap from Swift came in at 17.47 seconds, half a second faster than the fastest
03:12 human pilot.
03:13 A key advancement with Swift is that it can cope with real-world challenges, such as aerodynamic
03:19 turbulence, camera blur, and changes in illumination, which can confuse systems that attempt to
03:25 follow a pre-computed trajectory.
03:28 This victory is indeed a rare moment in the evolution of artificial intelligence, and
03:33 perhaps more such incidents would become common with time's ebb and flows.
03:38 you
Comments