Over recent years, artificial intelligence has showcased its exceptional capacity in tasks requiring intricate data processing, pattern discernment, and repetitive computation.
Such capabilities have allowed AI to make notable strides in diverse fields, often eclipsing human performance.
The AI-powered drone Swift AI, was designed by a team from the University of Zurich. In a competition involving speeds of up to 50mph (80 km/h) and intense accelerations that could cause blackouts in many humans, Swift AI emerged victorious in 15 of the 25 races.
Elia Kaufmann, one of the leading researchers behind Swift, proudly remarked, “Our result marks the first instance where an AI-powered robot has outperformed a human champion in a physically intensive sport crafted primarily for human competitors.”
Drone racing, viewed from a first-person perspective, necessitates deftly navigating drones through a series of gates.
Any misstep or error can lead to a crash. Pilots experience the race through a camera mounted on their drones, providing them a direct visual feed of the course.
Detailing their groundbreaking work in the esteemed journal Nature, Kaufmann, alongside his peers, elaborated on a series of competitive races between Swift and three top-tier drone racers: Thomas Bitmatta, Marvin Schapper, and Alex Vanover. The human contenders were given a week to familiarize themselves with the course to ensure a level playing field.
In contrast, Swift honed its skills in a digital realm, leveraging a virtual replica of the racing track.
Employing the advanced deep reinforcement learning technique, Swift deciphered the most efficient route around the circuit.
Though this trial-and-error-based method led to numerous crashes during the learning phase, the simulated environment meant a simple reset was all it took to get Swift back on track.
Elia Kaufmann, the study’s primary author, explained their meticulous approach: “To ensure the simulation closely mirrored real-world conditions, we devised a method to fine-tune the simulator using actual data.”
This accomplishment is a testament to AI’s boundless potential, pushing the boundaries of what is conceivable in sports and competition.
