SDN

该数据集包含了183个试验，总计8415条语音数据，大约18.7小时的控制流音频以及2.9小时的剪辑音频。这一数据集旨在评估智能体从人类那里预测对话行为，以及生成自身对话行为和物理导航动作的能力。此外，SDN基准测试旨在支持对自主驾驶智能体的情境交流进行实证研究，并能适应不可预期的事件。该任务主要聚焦于对话行为预测以及物理导航动作生成。

This dataset consists of 183 trials, totaling 8415 speech samples, with approximately 18.7 hours of control-flow audio and 2.9 hours of clipped audio. It is designed to evaluate the ability of AI Agents to predict conversational behaviors from human interactions, as well as generate their own conversational behaviors and physical navigation actions. Furthermore, the SDN benchmark is intended to support empirical research on contextual communication for autonomous driving AI Agents, and can adapt to unexpected events. This task primarily focuses on conversational behavior prediction and physical navigation action generation.