Future Drones: AI vs Human Pilots in Speed Showdown
Beyond the current limits of humanity, how fast can future drones be? Will artificial intelligence AI ultimately surpass humanity?
The organizer of the collaboration with Grepow, the University of Zurich Robotics and Perception Laboratory (UZH-RPG), has been studying AI-operated drones, and has invited pilots to collect and test data several times in the early stage.
This time, UZH-RPG Lab challenged three world champions, Thomas Bitmatta (2018 MultiGP International Open World Champion), Alex Vanover (2019 DRL World Champion), Marvin Schäpper (Swiss Drone League Champion), and at the same time, these three pilots are also TATTU pilots!

The drone pilots arrived at the competition venue a week in advance and continuously trained to familiarize themselves with the track. They also cooperated with the laboratory to collect eye movement data and other activities to help advance the speed of AI. The track and drones are specially made for the AI system. The 36 cameras and sensors are all designed to enable AI to accurately calculate the best route and obtain the fastest completion time.

World champion FPV drone pilot VS artificial intelligence AI competition
During this period, pilots probably had direct battles with AI about 2 to 3 times. Pilot Marvin Schäpper said, “I think on one lap human can be faster. But on 3 or more laps the AI will win. Cause they can do for example a 5.112s lap time 3 times behind each other cause they can just repeat one lap. (AI’s data algorithm is extremely powerful. A lap time of 5.112 seconds and repeating it three times is something humans cannot achieve.)”
Also in practice, the pilots are constantly adjusting their strategies and upgrading their equipment to achieve faster results. Flyer Thomas Bitmatta replaced the battery from the Tattu V4 1400mAh 6s to the V5, which is 0.194s faster!
”On the track we were doing, Thomas got down to a 6.26 on the Tattu 4.0, however on the 5.0 he managed a 6.066!!”
The night before the competition, that is, on the evening of June 10th. After the daytime practice, an internal test match not open to the public is quietly underway. The laboratory chose to hold the competition from 7 pm to 10 pm to test the AI system’s light recognition ability. However, due to Alex Vanover’s late arrival, the test match didn’t start until 9 pm. According to the postdoctoral members in the laboratory, the AI can reach a single lap in as fast as 4.3 seconds.
In the morning, the first round of the competition was launched. Pilots and AI Drones competed in one-on-one races and challenged for the fastest lap time in multiple attempts.
The order of the successive duels of the drone pilots is Alex Vanover, Thomas Bitmatta, and Marvin Schäpper. After multiple rounds of duels, Alex Vanover’s fastest time is 6.7 seconds, Thomas Bitmatta’s is 6.3 seconds, and Marvin Schäpper has reached 5.6 seconds. While the AI Drone took only 5.1 seconds for its fastest time in three duels.
But when competing against Alex, there was a small episode of “crashing the aircraft”.
In the first-round competition against pilot Alex Vanover, Alex came up with a method. His drone stayed in the middle of the crossing gate. The sensor of the AI Drone detected an obstacle and couldn’t pass through. The drone flew in another direction, resulting in a crash.
FPV (First Person View) Drone, abbreviated as FPV, is one of the drones with the highest operation difficulty coefficient on the market. The pilot can only use the first-person view and manually operate the flight through FPV glasses. Not only is the flight speed fast and the operation stability low, but the pilot can only see the state of high-speed movement in front of them. Therefore, it is very challenging for the pilot’s on-the-spot adaptability and operation skills.
In the racing field, the possibility of FPV collisions is very high. Pilots can only achieve functions such as bypassing obstacles and no-fly zones by appropriately adjusting the speed and attitude during maneuvering. Navigation, guidance, and control are all completed by artificial intelligence flight control, enabling drones with obstacle avoidance functions to fly on routes.
Therefore, in this high-speed flight state within a limited field of vision, the AI Drone needs to make judgments and adjustments within one hundredth of a second. Although it crashed, it successfully avoided a collision with Alex’s drone, reflecting the power of the drone’s algorithm from the side.
The second round was launched in the afternoon, mainly for image recognition. This time the AI changed the machine, from 36 sensor cameras to one camera on the head, which is equivalent to going from 36 brains to one brain. The results of this round have changed. Alex Vanover and Thomas Bitmatta are both 5.6S, Marvin 5.44s. Affected by light recognition, the AI slowed down, with a best score of 5.6S. The AI tried to fly again, but in the second after the host had just finished taking off smoothly, it directly crashed into the door by mistake. The battery was cut and started to catch fire and smoke. The organizers immediately evacuated the crowd. Christian, a member of the laboratory, rushed up immediately, picked up the battery and rushed out.
AI tried to fly again. But in the next second after the host just said that the takeoff was very smooth, it directly made a mistake and hit the door. The battery was cut and began to catch fire and smoke. The organizers immediately evacuated the crowd. Laboratory member Christian rushed up at the first time, picked up the battery and rushed out. The AI machine was taken to the laboratory for repair. For the time being, no repair plan can be found. And the previous four AI drones have all been damaged to varying degrees and cannot be used.
The competition was forced to be suspended. The professors held an emergency internal meeting to discuss. In order to give the audience a happy time, they decided to have another battle among the three pilots. In the pilot showdown, Alex Vanover won several consecutive matches. Excitedly, he ran towards the audience, waving and cheering.
The assistants checked the sensing system again, and the racetrack was rebuilt. It seems that the professors seem to have found a solution. Due to the previous battery fire incident, Grepow provided the organizer with a tattu V5 1200mah 6s battery with the same weight of 197g as a substitute for the previous Tattu V3 1800mAh 4s. However, considering the AI weight setting, the organizer did not use it in the end.
AI had another test flight and still couldn’t take off successfully. Fortunately, later, through the unremitting efforts of laboratory members, the AI system was finally repaired! The fourth round of the competition begins! Alex Vanover’s flight time is 5.288 seconds. Thomas Bitmatta’s is 5.48 seconds. The time after AI successfully takes off is 3.53 seconds!!!
Congratulations to the AI team! Applause to the human pilots! During the competition, the AI continuously adjusted the sensing system by observing the flight path and eye reaction of the pilots, and the pilots found the bug in the duel by observing the AI’s movement algorithm. Both are mutual achievements.



