Machine Protocols
How do machines that occupy physical space coordinate without centralized control?
Agent protocols solve coordination for digital agents — software talking to software. Machine protocols solve the harder version: agents that occupy physical space, consume energy, and collide when coordination fails.
The split isn't digital vs physical — it's retry-safe vs collision-unsafe. That failure mode asymmetry demands verified time, verified position, and fail-safe defaults that digital protocols don't need.
The five components of agency map directly to physical agents: foundations (power, connectivity), character (operational constraints, alignment), capabilities (sensors, actuators), capital (token stakes, data assets), drivers (objective functions, reward signals). Phygital beings that occupy physical space need all five — plus protocols that fail safe.
Protocol Bridge
Digital agents and physical agents follow the same pattern. The substrate differs.
| Function | Digital Agents | Physical Agents |
|---|---|---|
| Discovery | A2A Agent Cards | Peaq DID — machine publishes identity |
| Tool Access | MCP resources | Sensor APIs, edge compute |
| Commerce | UCP checkout | Peaq Data Marketplace |
| Payment | AP2 credentials | Peaq machine payments |
| Authorization | Verifiable Intent L1-L3 | Peaq DID + operational constraints |
| Coordination | A2A task delegation | Intercognitive orchestration |
| Time | System clocks (trusted) | UMT (verified on-chain) |
The key asymmetry: digital agents that lose connection retry. Physical agents that lose positioning collide. That asymmetry demands verified time, verified position, and fail-safe defaults.
IoT and DePIN
Machine-to-machine communication requires protocol surfaces that traditional networking does not address. Overview and advantages.
| Domain | Protocol Surface | Why It Matters |
|---|---|---|
| Positioning | RTK correction and attestation | Precision for robotics and autonomy |
| Connectivity | Wireless routing and throughput | Reliable data transport |
| Identity | Device identity and contribution proof | Trust and anti-sybil controls |
| Settlement | Payment and reward protocols | Incentive alignment |
| Interoperability | Cross-network handoff | Multi-system composition |
DePIN proved decentralized infrastructure scales faster than centralized alternatives. GEODNET built the world's largest RTK network in under two years using token incentives. The next step is standardizing how that infrastructure coordinates.
Dig Deeper
- Intercognitive Standard — Nine pillars for physical AI coordination: identity, fees, maps, sensors, positioning, compute, connectivity, orchestration, standards
- Peaq — Economy of Things — Machine identity, payments, data marketplace, and Universal Machine Time
Context
- Intelligent Hyperlinks — The pipe extends from screen to street when machines join the protocol
- Tokenization — Machines earn, hold, and trade tokenized value without human intermediaries
- Agent Protocols — Digital agent coordination (A2A, MCP, commerce)
- Verifiable Intent — Same delegation chain applies to machine authorization
- Agent Commerce — Standards war for autonomous transactions
- DePIN — Physical infrastructure incentivized by tokens
- Agent-Instrument Loop — How DePIN devices compose into feedback loops
- DePIN Devices — Hardware that earns while it works
- Robotics Industry — Where these protocols apply
- Agency — The five components physical agents need
- Phygital Beings — The actors these protocols serve
- Smart Contracts — Settlement layer for machine payments
- Protocols — Algorithms decide the route; protocols enable the handshake
Questions
What happens when the coordination protocol for physical agents fails — and the failure has physical consequences?
- Digital agents retry on disconnect. Physical agents that lose positioning collide. Does that asymmetry demand different protocol design, or just different failure modes?
- Can the same identity standard (Verifiable Intent + peaq DID) serve both digital and physical agents, or does embodiment require its own credential type?
- At what fleet size does centralized orchestration break, and what replaces it?