ASTM International 與新加坡標準理事會指出具身智能部署的五項標準化缺口

A robot can complete an impressive demonstration and still leave a prospective customer with difficult questions. How will it behave when people interrupt its task? Who controls the data its cameras collect? Can it work alongside machines from another supplier?

Those deployment questions are at the center of a white paper announced on November 3 by ASTM International and Enterprise Singapore, through the Singapore Standards Council (SSC). A Pathway for Future Embodied AI Standards Development identifies five standardization priorities for bringing intelligent physical systems into wider use.

Developed with Singapore’s Agency for Science, Technology and Research (A*STAR) and National Robotics Programme, the paper follows a July 2025 workshop involving more than 140 representatives from industry, government, research, and academia. Its agenda spans data governance, AI reliability, workplace safety, interoperability, and procurement. It is a starting point for standards development rather than a new certification that robot suppliers can already claim.

Why Embodied AI Changes the Deployment Question

Embodied AI connects machine intelligence to physical systems, including robots, autonomous vehicles, and other intelligent machines. It puts perception and decision-making into a setting where an error can affect people, equipment, and the surrounding environment.

As Unite.AI has explored in its coverage of how language models are changing embodied AI, increasingly capable AI opens new possibilities for robot behavior. The commercial question is how to make those capabilities dependable enough for everyday operations.

Consider a hypothetical delivery robot in a hospital. Completing a route is only one part of its job. It must also fit the building’s access arrangements, make its intentions understandable to staff, and operate within appropriate data-handling rules. A successful demonstration of navigation answers only part of a purchasing decision.

The Five Gaps Behind the Standards Agenda

The announcement groups the priorities into five areas. Their significance becomes clearer when viewed through the decisions a customer must make before deployment.

1. Data Governance and Privacy

The first priority is establishing common expectations for data governance and privacy compliance. A buyer needs to understand which information a system collects and how responsibilities are divided among the organizations involved.

For procurement teams, the practical implication is to make data arrangements part of the initial specification. In a hypothetical multi-vendor installation, an operator should be able to identify which supplier is responsible for each data flow without reconstructing the answer after an incident.

2. AI Safety and Reliability

The second area covers shared approaches to testing, benchmarking, and monitoring AI systems. Comparable evidence would give buyers a firmer basis for evaluating competing claims.

An important purchasing question follows: what exactly does a reported result describe? A test is easier to interpret when the task, operating conditions, software version, and failure criteria are explicit. Otherwise, two suppliers can report success while measuring substantially different things.

3. Workplace Safety and Risk Management

The third priority addresses physical safety and risks such as over-reliance on automation or poorly designed interfaces. Safe deployment therefore involves the people supervising and interacting with a machine.

A useful illustration is an alert that requires an operator to intervene. Its value depends on whether the person understands the message and has a workable way to respond. Installation planning should connect machine behavior, staff responsibilities, and intervention procedures into a coherent workflow.

4. Interoperability and Middleware

The fourth area concerns coordination among physical agents, software services, and infrastructure. For an enterprise assembling a fleet, the relevant unit of evaluation becomes the working installation.

Imagine buying two robots that can each navigate a corridor independently. The integration question is how they share constrained space and coordinate access to the same doorway. Purchasing each machine separately can leave that question unresolved until implementation. Explicit integration requirements give suppliers something concrete to demonstrate before acceptance.

5. Quality Certification and Procurement

The fifth priority is building trust that a system fits its intended purpose. Shared certification and procurement approaches could help customers turn broad promises into assessable requirements.

The strongest buying specifications connect a capability claim to a defined use case. Moving a particular load along an agreed route under specified conditions is a more useful acceptance target than a general statement that a robot is intelligent. That precision also makes disagreements about performance easier to resolve.

From a White Paper to Usable Standards

ASTM International and Enterprise Singapore say they will bring the identified gaps to their respective technical committees and global networks to help prioritize future work. Organizations with relevant deployment experience are invited to participate.

For robot developers, the opportunity is to contribute concrete evidence about where integration or evaluation becomes difficult. For buyers, it is to explain which unanswered questions prevent a pilot from becoming a repeatable purchase. Those perspectives can help standards work address the decisions that actually determine adoption.

The paper’s importance lies in bringing these deployment questions into one agenda. A capable machine still needs a convincing case for how it will operate within a customer’s environment. Shared standards could make that case easier to build, compare, and maintain as embodied AI moves into more workplaces.