Open humanoid data is still scarce. What are you training on?
The permissive options are few: Humanoid Everyday for real teleoperation, NVIDIA's GR00T simulation sets, and large egocentric human video as a proxy.
AgiBot World and the GR00T N1.6 weights carry noncommercial terms, which matters if you plan to ship. What mix are you using, and how do you handle the licensing?
Referenced repos
Indexed from Hugging Face and linked upstream, not hosted here.
Teleoperated humanoid demonstrations across more than 260 tasks in seven categories, with RGB, depth, LiDAR, tactile and IMU streams at 30 Hz, stored in LeRobot v2 format.
NVIDIA's simulated trajectory sets for post-training GR00T, covering bimanual Panda, Fourier GR1 humanoid tabletop, single-arm Panda kitchen, BEHAVIOR R1 Pro household and Unitree G1 tasks, each stored as a LeRobot-style folder.
Real dual-arm demonstrations recorded on a fleet of AgiBot robots across 36 household, retail and industrial tasks, shipped as tar archives with a script to convert them to LeRobot format.
About 10,000 hours of 1080p head-mounted video of factory workers doing hands-on jobs, delivered as roughly 193k H.265 clips (about 16 TB) with per-clip metadata.
A 3B cross-embodiment robot foundation model from NVIDIA that combines a vision-language backbone with a diffusion-transformer action head and can be post-trained on a new robot with a modest amount of data.