开源1万小时的具身智能数据？这家公司是为了什么

Core Viewpoint - The article discusses the release of the largest and most generalized open dataset for embodied intelligence, "10Kh RealOmni-Open DataSet," by JianZhi Robotics, which aims to accelerate the development of household robots through high-quality, real-world data [1][10]. Group 1: Dataset Overview - The dataset consists of over 1 million clips and more than 10,000 hours of embodied data, making it the largest in the industry [1][11]. - It focuses on 10 common household tasks, ensuring that each skill has over 10,000 clips, which provides extensive coverage for individual skills [3][5]. - The dataset is designed to be practical, emphasizing skill depth rather than merely expanding the number of skills [3]. Group 2: Data Quality and Specifications - The dataset features high-quality visuals with a resolution of 1600x1296 and a frame rate of 30 fps, recorded using a large FOV fisheye camera [4][5]. - It achieves centimeter-level trajectory precision, enhanced to sub-centimeter accuracy through high-precision IMU hardware and cloud reconstruction [4][13]. - The dataset includes diverse scenarios and natural human actions, collected from 3,000 real households, addressing the limitations of traditional data collection methods [8][10]. Group 3: Data Production and Automation - JianZhi Robotics has developed a complete data production chain, enabling rapid data collection and processing, accumulating nearly 1 million hours of data in just two months [10][14]. - The Gen DAS Gripper facilitates efficient data collection without the need for extensive site preparation, while the Gen Matrix platform ensures high-precision trajectory alignment and data quality [13][14]. - The Gen ADP automates the data annotation and processing workflow, allowing for continuous and rapid production of high-quality data [14]. Group 4: Importance of Open Data - The article emphasizes the necessity of open data to bridge the data gap, unify technical standards, and lower research and development barriers, ultimately accelerating the transition of embodied intelligence from laboratory settings to practical applications [16]. - The release of the "10Kh RealOmni-Open DataSet" is seen as a significant resource for innovation through data sharing, contributing to a positive feedback loop of data sharing, model optimization, and real-world application [16].