Austin VIOLA
收藏资源简介:
Austin VIOLA是一个用于机器人视觉操作的仿真与真实世界数据集,由德克萨斯大学奥斯汀分校与索尼人工智能公司合作创建。该数据集专注于机器人基于视觉的操控任务,包含多种日常操作场景,如桌面整理、咖啡制作等。数据集包含约100个仿真任务演示和50个真实世界任务演示,涵盖排序、堆叠、厨房操作等任务类型。数据来源于人类通过3D鼠标远程操作机器人的演示,包含RGB图像、机器人关节状态和末端执行器控制指令等多模态信息。创建过程中,通过颜色增强等技术增加视觉多样性,以提升策略的泛化能力。该数据集旨在支持机器人学习算法的研究,特别是针对复杂视觉环境下的操作任务。
Austin VIOLA is a simulation and real-world dataset for robotic vision-based manipulation, co-created by The University of Texas at Austin and Sony AI. This dataset focuses on vision-based robotic manipulation tasks, covering various daily operation scenarios such as tabletop organization, coffee preparation, and more. It contains approximately 100 simulated task demonstrations and 50 real-world task demonstrations, covering task types like sorting, stacking, kitchen operations, and others. The data is sourced from human teleoperated robotic demonstrations via a 3D mouse, and includes multimodal information such as RGB images, robot joint states, and end-effector control commands. During its development, techniques like color augmentation were adopted to increase visual diversity, thereby enhancing the generalization capability of learned policies. This dataset aims to support research on robotic learning algorithms, especially for manipulation tasks in complex visual environments.




