A Evolutionary Behavior Tree AI for Neural MMO Challenge

The Neural MMO Challenge aims to study robustness and teamwork in large-scale multi-agent environments. AI needs to explore, search and fight in large-scale environments, and get higher scores than other competitors. Therefore, we developed AI based on evolutionary behavior tree, which has the advantages of strong interpretability, low coupling between modules and strong robustness. Specifically, the AI adopts the idea of divide and conquer, divides the decision-making into two levels: team and individual, divides the four task achievements into several subtasks, finds appropriate agents for each subtask according to the current advantages, and then completes specific actions such as attack, foraging, avoidance and collaboration, and optimizes the parameters with evolutionary strategy algorithm.AI won the silver medal in IJCAI 2022 Neural MMO Challenge.