LightwheelAI, in partnership with Hugging Face, has released EgoSuite-Open100K, a dataset containing 100,000 hours of egocentric human activity. The first 10,000 hours are now available on the Hugging Face Hub, with the full dataset to be released in stages. This dataset includes 15,000+ tasks and 15,000+ real-world collection scenes, offering a comprehensive resource for Physical AI research and development. The dataset is designed to provide detailed supervision for robot training, capturing human actions such as reaching, grasping, and task completion. It also includes environmental categories and scene types to support diverse applications. LightwheelAI emphasized the importance of this dataset in addressing the challenge of scaling egocentric data for robotics, which is essential for training robots to perform complex tasks. The dataset is structured into multiple capture configurations, with varying levels of annotation depth to cater to different research and development needs. The release of EgoSuite-Open100K is part of an effort to establish shared standards for egocentric data, enabling better collaboration and comparison across the field. Source: huggingface
The EgoSuite-Open100K dataset includes 100,000 hours of first-person human activity across 15,000+ tasks and 15,000+ real-world collection scenes. It is structured into two main capture configurations: EgoStandard and EgoPro. EgoStandard includes standard egocentric capture, with sub-SKUs for hand pose and full body pose. EgoPro adds a wrist-mounted camera for close-range interaction, enhancing the capture of fine details and occluded movements. The dataset also includes three types of annotations: hand pose, body pose, and event-level semantic annotation on selected subsets. These annotations provide structured data for various applications, including action recognition and task understanding. The dataset is released in LeRobot v3 and MCAP formats to ensure compatibility with different robotics and multimodal data pipelines. LightwheelAI encourages the community to provide feedback on the dataset, which will help shape the next 90,000 hours of data. Source: huggingface
LightwheelAI and Hugging Face released EgoSuite-Open100K to address the challenge of scaling egocentric data for robotics. The dataset aims to provide detailed supervision for robot training, capturing human actions such as reaching, grasping, and task completion. It is designed to support a wide range of applications, including VLA model pretraining, human-to-robot behavior transfer, and egocentric representation learning. The release of this dataset is part of an effort to establish shared standards for egocentric data, enabling better collaboration and comparison across the field. Source: huggingface
