Pommerman Multi-Agent Training Dataset

该数据集采用了演员-评论家算法结合近端策略优化（PPO）的方法，在多智能体环境中训练智能体，通过课程学习和自我对弈的方式让智能体学会玩Pommerman游戏。数据集包含了采用不同策略的各种智能体，如静态智能体和基于规则的智能体，并使用Elo评级系统进行匹配和评估。该数据集规模涵盖了100轮循环赛的评估，任务是进行多智能体强化学习。

This dataset adopts the Actor-Critic algorithm combined with Proximal Policy Optimization (PPO) to train agents in a multi-agent environment, enabling the agents to master the Pommerman game via curriculum learning and self-play. The dataset includes various agents with different strategies, such as static agents and rule-based agents, and uses the Elo rating system for matchmaking and evaluation. This dataset covers evaluations of 100 round-robin matches, with the task of multi-agent reinforcement learning.