RT-1 Robot Action
收藏资源简介:
RT-1数据集由Google Robotics团队创建,旨在推动机器人在真实世界中的多任务学习与控制能力。该数据集包含约13万条机器人演示数据,涵盖超过700种不同的任务指令,涉及多种日常物品与操作场景。数据来源于17个月内13台机器人在厨房环境中的操作演示,包含丰富的视觉、语言和动作信息,总Token数达数百万级别。数据集的创建过程通过大规模的机器人自主操作与人类示范相结合,确保了数据的多样性和实用性。其主要应用于机器人学习领域,致力于解决机器人在复杂真实环境中对新任务、新物体和新环境的泛化能力问题,助力机器人实现高效、灵活的多任务操作。数据集可通过相关项目网站访问。
The RT-1 dataset was developed by the Google Robotics team, with the goal of advancing multi-task learning and control capabilities of robots in real-world environments. This dataset contains approximately 130,000 robot demonstration records, covering over 700 distinct task instructions and involving various daily objects and operation scenarios. The data is collected from the operational demonstrations of 13 robots deployed in kitchen settings over a 17-month period, and it encompasses rich visual, linguistic and kinematic information, with the total number of Tokens reaching millions. The dataset is constructed through a hybrid approach combining large-scale autonomous robot operations and human demonstrations, which ensures the diversity and practicality of the collected data. Its main applications are in the field of robotic learning, where it aims to solve the generalization challenges of robots when encountering new tasks, novel objects and unfamiliar environments in complex real-world scenarios, enabling robots to achieve efficient and flexible multi-task operations. The RT-1 dataset is accessible via the official project website.




