VRB Dataset

该数据集名为VRB，包含了从人类互动视频中收集的数据，旨在训练视觉可供性模型以支持机器人学习。通过这种模型，机器人能够估计人类在场景中可能互动的位置和方式，从而执行复杂任务。该数据集还包括了各种任务下的轨迹数据，其性能指标显示，与先前方法相比，机器人在学习方面的表现有显著提升。数据规模方面，该数据集从8个不同的环境收集数据，覆盖了多种任务。具体任务涉及在多样化环境中的模仿与探索式机器人学习。

This dataset, named VRB, contains data collected from human-interactive videos and is designed to train visual affordance models for robotic learning. Such models enable robots to estimate the locations and manners of human interactions in a scene, allowing them to execute complex tasks. The dataset also includes trajectory data across various tasks, and its performance metrics demonstrate that robots achieve significant improvements in learning performance compared to prior approaches. In terms of data scale, VRB collects data from 8 distinct environments and covers a wide range of tasks, with specific tasks involving imitation and exploratory robotic learning in diverse environments.