Downloads

Downloads & whitepaper

Technical documents from Maucher CNC-Robotic to download. Currently available: the whitepaper "Physical AI in industrial reality" – a trust infrastructure for moving humanoid and service robotics from demonstration into reliable, verifiable operation. (The whitepaper is in German.)

Whitepaper · PDF

Physical AI in industrial reality

A trust infrastructure for humanoid robotics, service robotics and reliable operation. A perspective from mechanical engineering, industrial manufacturing, robotics integration and CE-related risk assessment.

PDF · 17 pages · As of: June 2026 · Published by: Maucher CNC-Robotic GmbH · in German

Download whitepaper Open in browser
What it is about

From robot product to controlled operation

Physical AI describes AI systems that do not just process data but act in the physical world through sensors and actuators. In humanoid robots, mobile service robots and autonomous capture systems, artificial intelligence thus becomes a moving, grasping and perceiving technology. The central question is no longer only whether a robot can perform a task, but whether it may perform it responsibly in a specific environment.

The whitepaper describes how the gap between technical demonstration and reliable operation is closed: through a trust infrastructure made up of Operational Design Domain (ODD), real-world environment data, risk assessment, technical file, test cases, simulation, monitoring, cybersecurity, data protection, change management and a clear division of human responsibility. The existing deterministic safety logic is not replaced, but extended with a traceable operating logic.

The five core statements

Core of the whitepaper

  • Operation, not show — The technical demonstration of a humanoid robot is no proof of safe and lasting operation.
  • ODD as the foundation — Without a precise Operational Design Domain, no serious statement can be made about permitted use.
  • Real-world data instead of assumptions — Buildings, production areas and process environments must be measured, structured and versioned before they can serve as a basis for simulation and approval.
  • AURA ONE as an example — Real environments are made machine-readable. Other systems can serve the same purpose if they meet comparable evidence criteria.
  • Agent logic as a tool — AI agents improve structure, completeness and contradiction checking, but replace neither human responsibility nor a formal assessment body.
Key concepts

What the whitepaper explores in depth

Operational Design Domain

The contract between technology and reality

Which task, which spaces, which people, which loads, which speed and which abort criteria are approved – linked to real maps, zone models and version states.

Trust infrastructure

Verifiable operation instead of promises

Use-case description, real-world environment data, ODD, risk assessment, test cases, technical file, monitoring and governance together form the actually marketable product: controlled operation.

AURA ONE

Making real environments machine-readable

First measure, then structure, assess, test, approve and monitor. Capture does not produce a certification, but a reliable data basis for ODD, simulation and evidence.

Gaussian splats

Visualisation yes, safety proof no

Photorealistic 3D representations are valuable for communication and training, but no evidence of metric accuracy or collision safety. That requires verified geometry and semantic zones.

AI agents

Structuring review and documentation logic

A distributed agent system with clear roles reviews use cases from different perspectives, demands evidence and prepares decision templates – without replacing human responsibility.

Maturity model

From demonstration to auditable operation

Six stages from pure demonstration through the structured pilot to continuously auditable operation – each with its typical risks and revalidation triggers.

Glossary

Key terms

Physical AI — AI systems that act in the real world through sensors and actuators and thereby create physical effects.

Operational Design Domain (ODD) — The permitted operating domain of a system, with tasks, spaces, people, loads, conditions and limits.

Digital trust infrastructure — The totality of data, processes, evidence, roles and monitoring for the controlled use of Physical AI.

Near miss — An event in which no harm occurs, but a safety-relevant borderline case becomes visible.

Gaussian splat — A photorealistic 3D representation from image data, useful for visualisation but not automatically metrically or safety-technically reliable.

Revalidation — Renewed assessment after changes to the task, environment, software, data, sensors or operating conditions.

Technical file — Structured documentation of the relevant documents, evidence, assumptions, versions and approvals.

Frequently asked questions

Physical AI – briefly explained

Physical AI refers to AI systems that do not just process data but act in the physical world through sensors and actuators – for example in humanoid robots, mobile service robots and autonomous capture systems. What matters is not only whether a robot can perform a task, but whether it may perform it responsibly in a specific environment, with specific people and specific limits.
The Operational Design Domain (ODD) describes the permitted operating domain of a robot: for which task, in which environment, with which load, with which people, at which speed, with which supervision and with which abort criteria it may be deployed. It is the contract between technology and reality and prevents a creeping expansion of use without renewed review.
No. Gaussian splats (3D Gaussian splatting) are excellent for photorealistic visualisation of real spaces, but are not proof of metric accuracy or collision safety. Reliable robotics approvals additionally require metrically verified geometry, semantically validated zones, documented tolerances, collision models, navigation maps and a link to the ODD.
AURA ONE is an example from Maucher CNC-Robotic of an industrial capture and evidence logic: real operating environments are measured, structured, semantically annotated and versioned, so that reliable operating limits (ODD), simulations, test cases and verifiable evidence can be derived from them. Other capture systems can serve the same purpose if they meet comparable evidence criteria.
Maucher CNC-Robotic combines mechanical engineering, robotics integration and its own industrial manufacturing as part of the Maucher Group (a tier-1 supplier with three plants on Lake Constance). This gives rise to a trust infrastructure that moves Physical AI from the demo phase into reliable, verifiable operation – with ODD, real-world environment data, test cases, monitoring and clear responsibility.

Position Physical AI in your operation

Looking to move service or humanoid robotics from the demo phase into real-world use? We support you with use-case description, ODD, real environment capture and preparation of the evidence.

Discuss a project Physical AI & Sim-to-Real