FreeAskWorld Dataset
收藏资源简介:
FreeAskWorld是由清华大学人工智能产业研究院构建的大规模具身智能基准数据集,专为人类中心的社会导航研究设计。该数据集包含重构的城市场景、16类核心物体、63,429个标注样本帧及超过17小时的交互数据,涵盖对话历史、全景RGB图像和占据栅格图等多模态信息。数据集通过模块化生成流程创建,整合了人物模拟、场景随机化和语义指令生成技术,支持动态环境下的多轮交互。主要应用于视觉语言导航领域,旨在解决传统导航系统在动态目标适应、社会意图建模和实时人机交互方面的核心挑战。
FreeAskWorld is a large-scale embodied intelligence benchmark dataset developed by the Institute for Artificial Intelligence Industry Research of Tsinghua University, specifically designed for human-centric social navigation research. This dataset includes reconstructed urban scenes, 16 categories of core objects, 63,429 annotated sample frames and over 17 hours of interactive data, covering multimodal information such as dialogue histories, panoramic RGB images and occupancy grid maps. The dataset is constructed via a modular generation pipeline, integrating technologies including character simulation, scene randomization and semantic instruction generation, and supports multi-turn interactions in dynamic environments. It is primarily applied in the field of Vision-and-Language Navigation (VLN), aiming to address the core challenges faced by traditional navigation systems in dynamic target adaptation, social intention modeling and real-time human-computer interaction.

- 1FreeAskWorld: An Interactive and Closed-Loop Simulator for Human-Centric Embodied AI清华大学 · 2025年



