Collaborative museum heist with reinforcement learning

Non-playable characters (NPCs) play a crucial role in enhancing immersion video games. However, traditional NPC behaviors are often hard-coded using methods such as Finite State Machines, Decision and Behavior trees. This has few limitations; namely, it is quite difficult to implement complex cooperative secondly this makes easy for human players identify exploit patterns behavior. To overcome these challenges, Reinforcement learning (RL) can be used generate dynamic real-time responses player actions. In paper, we report on first results of applying RL techniques Non-Zero Sum, adversarial asymmetric game, multi-agent team. The game environment simulates museum heist, where the objective successfully trained team robbers with different skills (Locksmith, Technician) steal valuable items from without being detected by scripted security guards cameras. Both agents were concurrently separate policies received both individual group reward signals. Through training process, learned cooperate effectively use their maximize benefits. These demonstrate feasibility realizing full at same time achieve goals.