Multi-Agent Crafter

该数据集专为评估开放性、可扩展的协作任务中的战略协调、规划和资源共享而设计，尤其适用于大型语言模型驱动的多代理系统。该数据集特别有助于评估在多代理场景中基于大型语言模型的协作和记忆质量，解决了现有环境过于简单或过于复杂而无法进行有意义的协作的局限性。该数据集在代理数量上具有可扩展性（已测试1、2、6个代理），其任务是通过完成一系列分层的任务来收集一颗钻石。

This dataset is specifically designed to evaluate strategic coordination, planning, and resource sharing in open and scalable collaborative tasks, and is particularly tailored for large language model-driven multi-agent systems.
It is especially useful for assessing the collaboration and memory quality of LLM-based agents in multi-agent scenarios, addressing the limitations of existing environments that are either overly simplistic or overly complex to support meaningful collaboration.
This dataset supports scalability in the number of agents (1, 2, and 6 agents have been tested), with the agents' task being to collect a diamond by completing a series of hierarchical tasks.