The very first Flappy Bird AI Olympics took place on the 29th January 2020 during the exercise on Human Computer Interaction. Ten students let their AIs compete against each other to determine three winners.

Throughout the laboratories of the HCI exercise, the students did not only work with VR technologies and eye-tracking but also with artificial intelligence (AI). Besides solving classifications of text and pictures in jupyter notebooks, they also had the opportunity to train an agent in a competitive scenario, that implemented a Deep-Q-Network (DQN), to solve the videogame "Flappy Bird". Such a trained agent can exceed human capabilities very quickly.

The students subsequently created an observation- and reward-function for the agent and implemented a deep neuronal network with Python and Keras, that is the core of the DQN. The solutions ranged from a network that implemented only two layers with 16 neurons up to networks consisting of ten layers.

In a tense competition, the trained agents had the task to reach a score of 100. The size of the gaps, through which the bird has to fly, were subsequently lowered from 125px to 93px until three winners were determined. The places one to three go to:

1. Mr Benedikt Kentsch (Dense/4, Dense/24, Dense/48, 1.442 trainable params)
2. Mr Norman Reddig (3x Dense/88, 16.458 trainable params)
3. Mr Nils Verheyen (Dense/16, 66 trainable params)

All participants of the Olympics received a key fob, whilst the first three places were happy about a trophy made of acrylic glass.