BitRobot and Hugging Face Drop HIW-500: A Massive 10TB Real-Home Humanoid Dataset

BitRobot Network has released HIW-500, also known as Humanoids-in-the-Wild 500, described as the largest open-source humanoid teleoperation dataset collected in real household environments. The release was developed in partnership with Hugging Face and Unitree Robotics. The dataset captures more than 500 hours of whole-body motion across 23,000 episodes and more than 10 terabytes of raw data. It was gathered in 12 homes across Southeast Asia using the Unitree G1 humanoid platform and focuses on whole-body mobile manipulation tasks. Hugging Face's LeRobot team re-encoded the dataset, compressing it from approximately 10 terabytes to 2 terabytes with zero loss of fidelity. The data covers more than 10 core household tasks with detailed sub-task annotations. The full dataset is publicly available on Hugging Face in both native ROSbag and compressed LeRobot formats at huggingface.co/datasets/BitRobot/HIW-500-LeRobot.