Skip to content

RCAN Protocol

RCAN (Robot Communication & Authentication Network) is the open protocol that gives every robot a canonical identity, scoped permissions, and standardized message types.

Core concepts

Robot URI (RURI)

A canonical identifier for every robot, structured like a URL:

rcan://registry/manufacturer/model/version/serial

Example: rcan://opencastor.com/raspberrypi/pi5/1.0/RRN-000000000001

Robot Resource Name (RRN)

A unique identifier assigned at registration. Used in all telemetry, Firestore documents, and API calls.

RRN-000000000001 โ† Bob (Pi5 + Hailo-8L) RRN-000000000005 โ† Alex (Pi5)

Scopes

A hierarchy of access levels:

Scope Level Access
discover 1 Find the robot exists
status 1.5 Read telemetry and health
chat 2 Send commands, receive responses
contribute 2.5 Idle compute coordination
control 3 Motor control, physical actions
safety 4 Emergency stop, safety overrides

Message Types

36 standardized message types covering commands, telemetry, safety signals, contribute coordination, and authentication. See Message Types.

Safety invariants

RCAN encodes safety guarantees at the protocol level:

  • P66 โ€” contribute scope (2.5) always yields to control scope (3)
  • HMAC signing โ€” all safety-scope messages are HMAC-signed
  • Scope validation โ€” tokens are checked against scope before any action is taken

These are enforced in code, not configuration.

Recent protocol additions

See rcan.dev/compatibility for the full versioned feature matrix. Recent additions include:

  • ML-DSA-65 signing (FIPS 204) โ€” post-quantum algorithm replacing Ed25519
  • Multi-type entity numbering โ€” RRN (robots), RCN (components), RMN (models), RHN (harnesses)
  • LoA enforcement โ€” Level of Assurance gate on all control-scope commands
  • EU AI Act compliance โ€” firmware attestation (firmware_hash), SBOM publication, 10-year audit retention
  • Dual-brain architecture โ€” VLA reactive brain + LLM planning brain with confidence gate

Conformance is not certification.

Conformance to RCAN tracks (L1โ€“L4 protocol, Gateway Authority, HIL Runtime Safety) is self-asserted via signed bundles and independently replayable from those bundles. Conformance is not certification. Certification requires audit by a qualified third-party body, which is intentionally out-of-scope for the foundation in 2026.