机器人与物理 AI
ASTM 和新加坡标准理事会查明五个缺陷
在 Google 上的首选来源中添加 Unite.AI
机器人可以完成一个令人印象深刻的演示，仍然会给未来的客户留下难题。 当人们中断它的任务时，它会怎么做呢？是谁控制着摄像头收集的数据？能和另一个供应商的机器一起工作吗？
这些部署问题是11月3日ASTM国际和新加坡企业通过新加坡标准理事会(SSC)发表的一份白皮书宣布的中心。 《未来健全AI标准制定途径》确定了将智能物理系统更广泛地应用的五个标准化优先事项。
本文与新加坡科学、技术和研究机构(A*STAR)及国家机器人方案共同编写，是2025年7月由来自工业、政府、研究和学术界的140多名代表参加的讲习班之后编写的。 其议程涵盖数据治理、AI可靠性、工作场所安全、互操作性和采购。 这是制定标准的起点，而不是机器人供应商已经可以要求的新认证。
人工智能为何改变部署问题
健美AI将机器智能连接到物理系统，包括机器人，自主车辆，以及其他智能机器。 它将感知和决策置于一个错误可能影响人、设备和周围环境的环境下。
Unite.AI在报道语言模型如何改变包含的AI时探索，越来越有能力的AI为机器人行为开辟了新的可能性。 商业问题是，如何使这些能力在日常运作中具有足够的可靠性。
在医院里考虑假设的送货机器人完成一条路线只是其工作的一部分。 它还必须符合大楼的出入安排，使工作人员理解其意图，并在适当的数据处理规则范围内运作。 导航的成功展示只满足于购买决定的一部分。
标准议程背后的五个差距
公告将优先事项分为五个领域。 如果从客户在部署前必须作出的决定来看，其重要性就会更加明显。
1. 数据管理和隐私
第一个优先事项是确定对数据管理和遵守隐私的共同期望。 买方需要了解一个系统收集哪些信息以及有关组织之间如何划分责任。
对于采购小组来说，实际影响是使数据安排成为初步规格的一部分。 在假设的多供应商安装中，操作员应当能够确定由哪个供应商负责每个数据流，而不在事件发生后重建答案。
2. AI 安全与可靠性
第二个领域涉及测试、制定基准和监测人工智能系统的共同办法。 可比证据将为买方提供更坚实的评估竞争的声称的基础。
一个重要的购买问题如下：所报告的结果究竟说明了什么？当任务，操作条件，软件版本和故障标准明确时，测试更容易解释。 否则，两个供应商可以报告成功情况，同时衡量大不相同的情况。
3. 工作场所安全和风险管理
第三个优先事项涉及人身安全和诸如过度依赖自动化或设计不良的接口等风险。 因此，安全部署涉及监督和与机器互动的人员。
一个有用的例子是需要操作者介入的提示。 其价值取决于当事人是否理解信息，是否有可行的回应方式。 安装规划应将机器行为、工作人员的责任和干预程序联系起来，形成一个连贯的工作流程。
4. 互操作性和中间软件
第四个领域涉及实体agents、软件服务和基础设施之间的协调。 对于组建车队的企业，相关的评价单位成为工作设施。
想象一下，买下两个机器人每人可以独立导航走廊一体化问题是如何共享有限的空间和协调进入同一门道。 单独购买每台机器都可能使这一问题得不到解决，直至执行。 明确的一体化要求使供应商在接受之前有具体的证据。
5. 质量认证和采购
第五个优先事项是建立信任，使该系统符合其预期目的。 共同的认证和采购办法有助于客户将广泛的承诺转化为可评估的要求。
最强的购买规格将能力要求与界定的使用情况联系起来。 在特定条件下沿着约定的路线移动特定载荷，比一般地说机器人是智能的更有用的接受目标。 这种精确性也使得关于业绩的分歧更容易解决。
从白皮书到可用标准
ASTM国际和新加坡企业表示，它们将把已查明的差距带到各自的技术委员会和全球网络，以帮助确定今后工作的轻重缓急。 邀请具有相关部署经验的组织参加。
对于机器人开发者来说，机会是提供具体证据，说明集成或评价在何处变得困难。 对于买主来说，它是为了解释哪些未解问题使飞行员无法成为可重复的购买。 这些观点有助于标准工作解决实际决定收养的决定。
文件的重要性在于将这些部署问题纳入一个议程。 有能力的机器仍然需要一个令人信服的理由，说明它如何在顾客环境中运作。 共同的标准可以使这种情况更容易建立、比较和维持，因为体现的大赦国际进入更多的工作场所。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Robotics & Physical AI
ASTM and Singapore Standards Council Identify Five Gaps Holding Back Embodied AI
Add Unite.AI to your preferred sources on Google
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.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.