LLM+Debrief Simulation Environment

该数据集提供了一个类似健身房的模拟环境，用于评估大型语言模型代理在合作场景下使用自然语言交流的驾驶能力。该环境支持每个场景多达60个互动回合，并包含了在不同交流设置下的基线对比。数据集的规模涵盖了多种场景以及不同代理配置。任务目标是实现使用自然语言交流的协同驾驶。

This dataset offers a gym-like simulation environment for assessing the driving capabilities of large language model (LLM) agents that employ natural language communication in collaborative scenarios. This environment supports up to 60 interaction rounds per scenario, and incorporates baseline comparisons across different communication settings. The dataset encompasses a wide variety of scenarios and diverse agent configurations. The task goal is to achieve collaborative driving through natural language communication.