TEG Report: From Automation to Economy - Why Manufacturing Needs a Common Autonomy Framework
Background
Manufacturing is undergoing a profound shift driven by AI, robotics, and cyber-physical production systems. Machines increasingly sense, interpret, and act on their own — capabilities widely lumped together under "autonomy." Yet unlike fields such as autonomous driving, medical technology, or aviation, manufacturing still lacks a consistent, widely accepted definition of autonomy. This position paper — authored by the Artificial Intelligence Application Technical Expert Group under the Sino-German Standardisation Cooperation — argues why that needs to change.
Lessons from Mature Sectors
Three domains show how autonomy can be formalized without stifling innovation:
- Autonomous driving: A shared taxonomy defines levels of automation, operational design domains, and human-machine responsibility handovers — used by engineers, regulators, and insurers alike.
- Medical technology: Autonomy is regulated through risk- and responsibility-based models rather than fixed levels, with strict rules on when systems may act, be overridden, or require human oversight.
- Aviation: Graduated authority and rigorous certification ensure autonomous functions remain bounded, fail-safe, and tightly coupled to human factors.
Across all three, autonomy is treated as a bounded allocation of decision-making between humans and machines — never unconstrained independence.
The Gap in Manufacturing
Manufacturing lacks this structure. Terms like "automated," "autonomous," "intelligent," and "self-optimizing" are used interchangeably, without distinguishing decision scope or human involvement. Regulation remains fragmented, addressing safety and function in isolated silos rather than treating autonomy as a systematic, scalable attribute. This is increasingly problematic as:
- Distributed, agent-based, and AI-enabled architectures replace centralized control
- Humanoid robots and foundation models (LLMs, vision-language-action models) introduce learning-enabled, emergent behavior that defies traditional deterministic assessment
- Standardization bodies (robotics, automation/control, AI) each apply their own implicit assumptions, creating inconsistent and hard-to-compare requirements
Consequences of Inaction
The absence of a shared framework creates concrete problems: unclear liability when AI-driven decisions fail, difficulty comparing systems with similar functionality but different autonomy profiles, operational risks from mismatched training and oversight practices, and — especially relevant for Germany and China as integrated manufacturing partners — divergent regulatory interpretations that complicate cross-border deployment and certification.
The Proposal
The paper calls for a dedicated, internationally accepted standard on autonomy in manufacturing — technology-agnostic, focused on decision authority, human involvement, operational boundaries, and system behavior rather than mandating specific architectures. Concretely, it recommends:
- Launching a New Work Item Proposal, ideally anchored within IEC TC 65 WG31 (industrial agents and multi-agent systems)
- Close, ongoing exchange with the research community to translate concepts (autonomy levels, explainability, human oversight, verification/validation) into standardization language
- Establishing an international cross-sector round table linking manufacturing, transportation, and healthcare standardization communities
Why It Matters for Germany and China
Both countries are frontrunners in intelligent manufacturing and face the same rising complexity as autonomy scales from machine to factory level. A jointly supported framework would reduce regulatory friction, enable smoother cross-border collaboration, and let both countries help shape emerging international standards proactively rather than reactively.