Artificial intelligence (AI) is once again taking on humans, this time in drone racing. The relevant paper was published in the British magazine Nature on the 30th.
Researchers from the University of Zurich in Switzerland and other institutions have developed an artificial intelligence system called Swift based on deep reinforcement learning. The system's ability to control drones can defeat top human players in one-on-one competitions.
Artificial intelligence has defeated humans in game duels such as chess and "StarCraft", but these are chess or video games, not real-world competitions. In this first-person perspective drone racing competition between artificial intelligence and humans, contestants need to control a drone flying at high speed (up to more than 100 kilometers per hour) through a three-dimensional track, and the entire process is captured by the video of the onboard camera. Live broadcasting, observing and controlling the surrounding environment from the perspective of a drone, is more unpredictable than chess and video games, and is not easy for artificial intelligence.
According to reports, this artificial intelligence system combines deep reinforcement learning strategies with data collected from the real world, and defeated three human players who had won the first-person perspective drone racing competition. He won 15 out of 25 games and broke the record for the shortest time in the game, half a second ahead of the best time of a human player.
The research team stated that this is a milestone in the development of artificial intelligence and will contribute to the research and development of autonomous vehicles, robots and other related fields.
