Reinforcement Learning Applications in Gamified Scenarios

Solution development using ML algorithms became a trend in various fields. As a consequence, it rapidly pushes most of these fields forward in the last decade. The game development was not left behind. Moreover, games are frequently used to benchmark AI algorithms; various open-source environments are built for researchers to evaluate their algorithms there consistently. The algorithms of NPCs possession are becoming believable, natural for players. The project covered the latest and well-known AI techniques in games. Also, it incorporates the gamified scenario similar to "Musical Chairs" game design. Agents learned to play using Reinforcement Learning, Proximal Policy Optimisation algorithm. Agents had various sensors—sphere rays and camera sensors. Agents acted in both discrete and continuous spaces. Training of all models in the last setup was bounded in 10^6 episodes. All the results were gathered, compared and discussed.