RobotSlang
收藏资源简介:
RobotSlang数据集由密歇根大学创建,包含169个自然语言对话,用于研究机器人定位和导航。数据集包含近5000条语音和超过1000分钟的机器人摄像头和控制流。该数据集适用于训练语言引导的机器人遵循人类语言指令,解决从搜索救援到辅助指导等多种应用场景中的定位和导航问题。数据集的创建涉及同步、捕获和注释原始对话文本、传感器观测和地图数据,以创建RobotSlang基准。
The RobotSlang dataset, developed by the University of Michigan, contains 169 natural language dialogues for research on robot localization and navigation. It includes nearly 5,000 speech segments and over 1,000 minutes of robot camera footage and control flow data. This dataset is suitable for training language-guided robots to follow human language instructions, solving localization and navigation problems across various application scenarios ranging from search and rescue to assistive guidance. The creation of the RobotSlang benchmark involves synchronizing, capturing, and annotating raw dialogue texts, sensor observations and map data.

- 1The RobotSlang Benchmark: Dialog-guided Robot Localization and Navigation密歇根大学 · 2020年



