Research reference · source and simulation evidence
A Physical AI research reference for shared-resource coordination.
Shared-resource decisions across device, edge, and cloud boundaries.
A neutral horizontal runtime and coordination substrate for physical agents across device, site, edge, and cloud boundaries. The reviewed record establishes source and simulation scope for a governed horizontal candidate and selected vertical packs; it does not establish physical deployment.
Documented research reference
Source through 2026-08-24
Inspectable artefacts
Explicit claim boundary
Business ↔ technology bridge · Physical AI reference implementation
Pressure-tests product and architecture decisions against a real cross-domain Physical AI source system.
This is a portfolio and source interpretation—not a customer, deployment, certification, or outcome claim.
Every new physical-agent vertical can become another bespoke platform, multiplying integration, operations, support, and assurance cost.
Separate the reusable horizontal coordination substrate from vertical protocol ownership and keep final safety authority with the local controller.
Signed proposals, resource leases, policy and admission boundaries, local veto, authoritative readback, catalogues, conformance paths, and cross-domain simulation.
Runtime constraints and validation gaps reopen platform scope, supported-vertical claims, deployment obligations, release decisions, and product economics.
Source evidence
A neutral horizontal runtime and coordination substrate for physical agents across device, site, edge, and cloud boundaries. The reviewed record establishes source and simulation scope for a governed horizontal candidate and selected vertical packs; it does not establish physical deployment.
Repository evidence is not client, deployment, certification, or outcome evidence.
The Problem
The system decisions a cross-domain coordination layer must make explicit
EV charging, robotics, energy, and infrastructure each bring their own protocols and authority models. A cross-domain site still needs an explicit rule for allocating shared power, space, time, and priority without erasing those vertical boundaries.
If a safety or resource decision depends on a cloud round-trip, the design needs a defined degraded mode. Auralink therefore treats edge continuity and local safety authority as architecture requirements to validate, not as already-proven uptime claims.
A site can contain multiple vendor stacks with different identities, telemetry, update policies, and operating constraints. Coordination has to preserve those ownership boundaries while making the shared-resource decision traceable.
Priority and pre-emption rules must distinguish routine work from safety and emergency authority. The source models that vocabulary; the response timing and physical effect still require hardware and site validation.
How It Works
A documented device, edge, and cloud split—with local safety authority kept explicit
Evidence
What the reviewed source establishes—and where another validation gate is still required
Engineering Discipline
Only a currently authorized, qualified research result appears here. Private inventory stays private.
Snapshot unavailable
The public facts require re-verification before publication.
Only three paths are exercised together in the retained golden-path simulation. The remaining entries are documented design scope, not deployment evidence.
A simulated EV-charging path exercises the generic coordination contract. Current OCPP and OCPI interoperability, real equipment, and certification remain separate work.
A simulated robotics path exercises workspace and power coordination. Open-RMF and strict VDA 5050 compatibility, real robots, and safety validation are not established by that simulation.
A simulated smart-city path exercises the generic coordinator. Current NTCIP/SNMP interoperability, controller hardware, and municipal deployment are not asserted.
Battery, solar, and site-load coordination appears as documented design and simulation scope. No live BESS, inverter, or energy-site validation is asserted.
Industrial protocol and scheduling boundaries are documented. Current OPC UA/Modbus interoperability, machine integration, and plant validation remain open gates.
HVAC and building-load composition is a documented application boundary. No building-management-system integration or site result is asserted.
ACS v0.1 at a glance
The Auralink Coordination Specification (ACS v0.1) proposes a small vocabulary for a reference coordinator and its simulated participants. Adoption by a device or protocol is not implied.
RESOURCEWatts on a substation, a busbar, a feeder, or a single charger contactor. Claims compose across hierarchical buses with hard physical caps.
RESOURCECubic metres on a factory floor or square metres at an intersection. The primitive for robot AGV reservations and traffic preemption.
RESOURCETime windows on a traffic signal head. The primitive for emergency-vehicle preemption, transit-priority requests, and demand-responsive timing.
RESOURCERF channel + bandwidth + time. The primitive for shared-radio environments — farms, ports, megaprojects with mixed 5G/LoRa/Wi-Fi.
RESOURCEDiscrete scheduled intervals on a shared physical asset (production cell, robot arm, dock). Composable with Power for energy-aware schedules.
RESOURCEBookkeeping primitive for OCPI roaming, virtual energy credits, V2G settlement, carbon allocation. The 'soft' resource type.
CLASS 1Represents hardware ESTOPs, backup loads, and other local emergency authority. Timing and physical effect require device-specific validation.
CLASS 2Represents safety-critical constraints such as thermal, collision, or grid-stability holds. The specification defines priority semantics; real safety behaviour is not established by source alone.
CLASS 3Represents bounded priority work such as service windows or charging schedules. Taper semantics are part of the reference design.
CLASS 4Represents deferrable work such as background charging or idle reservations. The reference policy makes this the first class considered for tapering.
Supply chain trust
Image signing
Signing and admission-control configuration is represented in source. Hosted workflow execution, image publication, and cluster enforcement are not asserted here.
Provenance + SBOM
Provenance, SBOM, and verification-script surfaces are present. A published attestation set must be checked separately for any specific release.
Audit chain
The source contains chained-audit and witness-anchor concepts and tests. Persistence, external anchoring, and operational retention require deployment validation.
Key custody
The signer and key-custody boundaries are designed to be pluggable. No particular TPM, KMS, or Vault deployment is claimed.
Transport
mTLS configuration and observability surfaces exist in source. Certificate lifecycle and live transport enforcement remain operator responsibilities.
Secrets
Credential-provider interfaces and Kubernetes secret templates exist. This does not establish a live Vault, ESO, or production cluster.
Deep Dive
Deep-dive into Auralink. NDA required for access.
These documents contain proprietary architecture details, benchmarks, and implementation specifics. A quick NDA protects both parties.
Why a coordination plane is the missing primitive for distributed physical AI, and what it is designed to enable across EV / robotics / smart-city / IIoT.
ACS v0.1 — six proposed normative resource types, four safety classes, graceful-taper preemption, federated learning under differential privacy. With the TLA+ formal model.
Full system architecture across cloud and edge, charger OS in Rust, agent bundle in Go, deployment topology, RAG backbone, and megaproject playbooks as documented design scope.
Source Code
View the GitHub repositories and HuggingFace models. Requires NDA acceptance first.
Please sign the NDA first before requesting code access.
Partnership
OCPP 2.0.1 + 1.6J represented in source, ROS 2 bridge prototyped, OpenADR 2.0b + IEEE 2030.5 in progress — current-protocol interoperability remains a separate gate. Looking for partners across the documented domains.
OCPP 2.0.1 / 1.6J, OCPI 2.2.1 roaming, and an ISO 15118-2/-20 framework are represented in source. The EV-charging path is exercised in simulation; current protocol interoperability, real equipment, and certification remain separate work with CPO partners.
ROS 2 bridge prototyped 2026-05 with stale-pose / failsafe / e-stop. OpenADR 2.0b + IEEE 2030.5 demos. Megaproject playbooks under apps/city/ for large-scale mixed-use and hyper-scale industrial sites are documented design scope, not deployment evidence.
The coordination spec (ACS v0.1) is a proposed normative vocabulary; breaking changes are meant to go through an RFC. Build your own coordinator, embed the SDKs, or run the reference implementation. The design aims to avoid lock-in to one implementation.
Roadmap
The next evidence gates for a Physical AI research reference.
The reviewed snapshot contains a governed component register, selected-suite transcript, architecture and specification material, formal-analysis artefacts, and a cross-vertical golden-path simulation. It is not a production verdict.
Validate production HAL boundaries, real device behaviour, current protocol adapters, fault injection, recovery, and timing against explicit acceptance criteria.
Choose one narrowly defined environment, document system and safety ownership, run acceptance and recovery tests, and retain evidence without generalising beyond that boundary.
Only after a bounded trial: test whether architecture, support, economics, safety evidence, and operating procedures repeat across another site or vertical. Advanced research remains separate from that scale decision.
Building production physical-AI infrastructure? Hyperion consulting covers pilot validation, production hardening, EU AI Act compliance, and full integration engagements.
Twelve weeks to a production-grade multi-agent system that serves as the software and control-plane complement to your cyber-physical stack — fleet intelligence, SCADA-adjacent orchestration, or autonomous operations — with the eval harness, the observability stack, and the SRE handoff your team needs to operate it
Twelve weeks to harden an edge or embedded AI pilot stuck before production — on constrained hardware, inside safety envelopes, under latency and reliability requirements the pilot was never designed to meet
From demo to dependable operation. Subscribe for new Physical AI reading.
Unsubscribe anytime. No spam, ever.