Sensor Modalities
Architecture
Engineered specifically for frontier humanoid robotics, embodied foundation models (Open X-Embodiment, RT-2, Octo), and spatial world models.
Pre-QC Raw Monocular Cam Video
Ultra-high-density head-mounted 4K / 1080p 60fps video capture logged from the egocentric perspective of skilled workers. Designed to provide high-frequency visuo-motor tokens for dexterous bimanual manipulation, tool engagement dynamics, and deformable object handling (textiles, ceramics, wiring, assembly).
Stereo Cam 3D Spatial Video
Dual-lens hardware-genlocked stereoscopic depth video capturing true 3D spatial geometry, binocular disparity maps, and sub-millimeter contact dynamics. Calibrated with precise factory intrinsic and extrinsic parameters for direct point-cloud reconstruction and spatial scene graphs.
AI Narrated Egocentric Video
Time-aligned synthetic audio narration combined with frame-by-frame Chain-of-Thought (CoT) action reasoning logs. Every video includes high-density text tokens articulating intentions, pre-contact subgoals, tool affordances, and contingency recovery strategies.
Human Narrated Video
Authentic, native voice commentary from master operators, artisans, surgeons, and industrial machinists. Captures unscripted, natural domain-expert cognition, safety caveats, tactile feedback descriptions, and non-verbal tool acoustic cues.