Graviti captures, annotates, verifies, and licenses production-grade training datasets for Physical AI—including humanoids, robotics, and world models—and large language models
Graviti captures, annotates, verifies, and licenses production-grade training datasets for Physical AI—including humanoids, robotics, and world models—and large language models. Its datasets include egocentric video, robot teleoperation trajectories, expert-written supervised fine-tuning data, human preference pairs, and agentic tool-use trajectories, with consent and provenance tracked for each record.