OpenAI, an artificial intelligence research company has recently done some interesting research by making artificial intelligent agents or bots to play hide-and-seek. The goal of this research was to observe multi-agent interaction between hiders and seekers (both as agents) and discover how these agents search and use available digital tools or resources inside a given virtual environment.
The rules of the game were simple. Each AI team had between one and three agents, and games lasted in 240 moves. The game consists of two competitors: hiders and seekers. Hiders consist of a team of bots that can hide by locking the object so that other bots cannot move it. Seekers consist of a team of bots that can search the hiders using available objects and other resources present. Seekers were frozen in place for the first 96 moves, giving hiders a chance to hide. Each agent was programmed to maximize its point totals, and points were awarded to all the members of a team at once.
The virtual environment consists of a square arena with boxes, ramps, and walls that agents could push around, lock and unlock it. The arena was based in a 3D physics simulator that prevented real-world impossibilities like agents walking through walls or two agents occupying the same space. The agents were trained to be competent in this game by playing millions of hide and seek games in virtual arenas. OpenAI says this self-play as the application of reinforcement learning where suitable action is taken to maximize reward in a particular situation.
During the research some interesting actions of bots were observed like bots discovered using ramp objects, surfing through objects and other technical flaws in the game environment to achieve their goal. Take a look at some surprising clips of these bots playing from the video below:
This amazing research on the multi-agent system can find useful applications in computer games, smart grids, transportations, networking and many more real-world applications making human life easier than before.