Sažetak
Video games in which human players control various vehicles to achieve specific goals can serve as a potential simulation environment for tracking and learning human control patterns. This study proposes a conceptual framework that can utilize video games as a training environment for the Artificial Intelligence (AI) based bots that control in-game vehicles in interactions with human players. The concept uses trained AI models of bots that interact with human players during the initial training steps for autonomous vehicles for transport and combat tasks. The transferred models from in-game AI bots can greatly stabilize learning convergence and speed up the overall training of real-world autonomous vehicles. Thus, this study addresses the problem of accountability and possible control mechanisms in generating desirable behavioural patterns of human players in controlling in-game transport and combat vehicles. Special focus is set on cognitive distortions within players' behavioral patterns, which can hinder their accountability for transfer learning.
Ključne riječi
Transfer learning Reinforcement learning Autonomous vehicles Video games Autonomous combat systems
