Humyn Labs develops data infrastructure and training datasets for robotics and physical AI, turning human activity in real-world environments into synchronized, multisensory signals. Its platform brings together sound, sight, mobility, and touch to help physical AI systems learn from richer experiences than visual data alone.

The company offers four data modalities. Its established voice and audio collection includes 50,000 hours across 33 languages, featuring in-environment recordings, rare dialects, and code-switching. Its sight offering is being built around 10,000 hours from more than 20 countries, with 6DoF pose, hand-skeletal data, IMU/SLAM, and episode-level annotation. Humyn is also integrating mobility data, including robotic-arm kinematics, whole-body pose, joint torque, and spatial mapping. Its touch work explores force, resistance, and compliance through sensor-equipped teleoperated grippers.

Humyn’s services span contributor sourcing, capture, synchronization, validation, processing, quality control, labeling, and benchmark evaluation. It provides traceable data with contributor and collection metadata, and delivers training-ready datasets in MCAP, RLDS, and LeRobot v3 formats. The company also offers sample datasets and collection scoping for prospective partners.