Skip to main content
Interoperability dies in the details, so Layer 3 pins them down:
  • Units — SI everywhere, encoded in field names (_m, _rad, _mps, _n). No implicit unit negotiation. See units and frames.
  • Coordinate frames — right-handed, with an explicit frame tree (named frames and parent transforms) published at session.ready, borrowed conceptually from ROS tf. Every pose names its frame.
  • Time — monotonic nanosecond timestamps against a session clock, anchored to wall-clock and (in sims) to sim-time. See timestamps and clocks.
  • Modalities — a registry of standard encodings (image/jpeg, audio/pcm, proprio/json, text/event+json, …) with x-vendor.* extension naming. See modalities.
  • Scene graphs — an optional semantic channel exposing entities, affordances, and relations as structured JSON. Enormously valuable for LLM agents that reason in symbols rather than pixels. See scene graphs.