Screenshot

TeamKill is an Artificial Intelligence Malmo-based project developed for a multiplayer Minecraft survival minigame “Falling Floor”.

Falling Floor

In this Minecraft minigame, the players’ mission is to survive by running around the map while floors start to crumble beneath them. This zero-sum game requires players to be the last one standing in order to win the game. The players are not allowed to jump or do any other actions other than to run for survival and employ various tactics to survive longer than their opponents. The single player version of this game requires player to survive as long as possible on the map under the same condition. Score is calculated based on the player’s survival time.

Our Agent

Our AI agent will try to survive in this environment by using our search and learn algorithm. The search part of the algorithm ensures agent’s survival by preventing self-entrapment in small isolated lands and providing the best direction the agent can move based on tile placement. Our learning algorithm allows the agent to learn and adapt to various, ever-changing situations.

Breadth First Search

Future Applications

Our algorithm employs search and reinforcement learning to estimate the best movement direction for our agent under a harshly competitive environment with a huge time constraint. When applied to a 3D environment, our search and learn algorithm can help agents to explore and survive in harsh, changing environments with unpredictable external agents such as debris-filled areas in the outer space and the deep ocean.