What if an autonomous drone powered by artificial intelligence took on the best human pilots in an epic battle in the air? This is the question posed by a group of researchers from the University of Zurich, who decided to test their AI system in an extreme sport: drone racing. The result was surprising: the autonomous drone not only managed to complete the circuit without crashing, but also surpassed the world champions of this discipline in speed and precision.
AI-powered drone beats top pilots
Artificial intelligence (AI) is not only capable of outperforming humans in board games, video games or chess, but also in extreme sports such as drone racing. This is demonstrated by a study published in the journal Nature, in which an AI system called Swift managed to beat three world champions of this discipline on a circuit full of obstacles.
Swift is an autonomous drone that was developed by a team of researchers at the University of Zurich (Switzerland), led by Professor Davide Scaramuzza. The system uses a deep reinforcement learning technique, which combines the processing of a large amount of data with the observation of rules that reward the machine’s progress.
The drone was trained in a simulated environment containing a virtual replica of the real circuit, where it crashed hundreds of times until it learned to avoid collisions. He then faced off against three expert human pilots, who wore helmets that transmitted images from the drone they were piloting.
Swift won 15 of 25 races against each of the human drivers and completed the fastest lap of the circuit, a half-second lead over the best time recorded by a person. Additionally, he made tighter and more precise turns than his rivals.
This is how this drone managed to win over the best pilots
And how was he able to achieve it? According to The Guardian, before the event, the human pilots had a week of practice, while Swift underwent training in a simulated environment that virtually replicated the competition. Furthermore, it is mentioned that “the drone experienced multiple crashes during its training phase” but since this was a simulation, the researchers simply restarted the process.
This is a neural network flying a drone at extremely high speed, beating human champions in FPV drone racing.
– Reinforcement learning as a tool is so marvelously versatile. It’s able to solve both fast, reactive tasks and slow, deliberate tasks (ChatGPT RLHF).
– Trained in… pic.twitter.com/gffbnGZTNX
— Jim Fan (@DrJimFan) August 31, 2023
When testing it in races, and to prevent collisions, Swift used a built-in camera that sent videos to a neural network responsible for detecting race gates. In this way, the researchers determined the commands necessary to calculate the position, speed and orientation of the drone.
The researchers highlight that this is the first case of a mobile autonomous robot achieving world champion-level performance in a real-world competitive sport. They also point out that the system may have practical applications in the development of autonomous vehicles, personal robots or unmanned aircraft.
More articles related to Curiosities
The latest in Curiosities
#autonomous #drone #powered #artificial #intelligence #defeats #pilots #epic #battle #air