Intent → Resource Resolution → Physical Actuation
1Physical AI Requires Physical Infrastructure Abstractions
Artificial intelligence has historically operated primarily within informational domains. Contemporary Physical AI extends intelligence into systems that must perceive physical state, reason under real-world constraints, and produce consequential actions in physical environments. This direction builds upon a substantial body of work in cyber-physical systems, embodied intelligence, robotics, edge computing, and intelligent infrastructure. The National Institute of Standards and Technology describes cyber-physical systems as systems integrating digital, physical, and human components through engineered interactions between computation and physical processes [1].
More recent work increasingly frames Physical AI around the ability of computational systems not merely to represent the world, but to perceive, act, and adapt within it [2]. As these capabilities mature, a second-order architectural problem emerges. Physical intelligence does not only require perception and action. It also requires mechanisms for determining what in the physical environment should be acted upon. Consider a person who requires a particular clinical supply, manufacturing component, or warehouse item. Conventional environments implicitly require the requester to participate in physical resolution. The requester must know, remember, infer, or discover where the desired resource is located.
In effect, human spatial knowledge becomes part of the system architecture. IAPI proposes that this responsibility can instead be assigned to infrastructure.
2Intent as the Request; Infrastructure as the Resolver
The requester specifies what is needed. The infrastructure determines where it is.
The central abstraction
This creates a separation between two concepts that are commonly coupled in physical environments: resource identity and resource location. Under conventional physical retrieval, a request such as "I need an IV catheter" is incomplete from the standpoint of physical execution. The user must subsequently determine which room, cabinet, rack, shelf, or slot contains the required item. Under IAPI, location is not required to be part of the user's request. Instead, the infrastructure receives or infers an intent, evaluates available physical resources against that intent and the current system state, selects an appropriate resource, resolves its present physical endpoint, and generates guidance or actuation.
The architectural sequence is therefore WHAT → WHERE → GUIDANCE / ACTUATION or, more generally, Intent → Resource Resolution → Physical Actuation. The significant transition is not simply from slower search to faster search. It is from human-executed search to infrastructure-executed resolution.
3IAPI as a Layer Within Physical AI
IAPI is not intended as a synonym for Physical AI. Physical AI is the broader technological framework. It includes systems capable of sensing, interpreting, deciding, learning, adapting, and acting within physical environments. These systems may include robots, autonomous vehicles, intelligent manufacturing systems, medical devices, smart buildings, and other cyber-physical systems. IAPI addresses one narrower architectural question within that framework: how is intent resolved against a changing population of physical resources?
| Layer | Scope |
|---|---|
| Physical AI | Systems that perceive, reason, and act in the physical world. |
| Intent-Addressable Physical Infrastructure | An infrastructure abstraction for resolving expressed intent to physical resources: Intent → Resource Resolution → Physical Actuation. |
This distinction matters because sophisticated physical intelligence alone does not eliminate the need for resource-resolution abstractions. A system may possess highly capable perception, reasoning, or robotic actuation while still requiring explicit knowledge of which physical endpoint should be addressed. IAPI places that resolution function inside the infrastructure. The requesting entity may be a person, a workflow system, or a machine agent. The abstraction remains the same: the requester specifies the required resource or outcome; the infrastructure resolves the appropriate physical endpoint.
4From Location Addressing to Intent Addressing
Modern computing relies heavily on abstraction between requested resources and their underlying locations. Applications routinely request information without knowing the physical memory cells, storage devices, processors, or network paths involved in satisfying the request. Information-centric networking similarly explores architectures in which communication is organized around named information rather than requiring applications to reason directly about the location of the host containing it [3]. IAPI applies a related architectural principle to physical resources.
The analogy is not exact. Physical objects differ fundamentally from digital information. They are spatially constrained, rivalrous, state-dependent, and often movable. A catheter cannot be replicated or routed like a packet. Those differences make physical resource resolution a distinct problem. Nevertheless, the abstraction remains useful: the identity of the requested resource does not need to be identical to its physical locator. A physical resource can therefore be treated as a resolvable endpoint whose location is determined when the request is evaluated rather than encoded permanently into the request itself. This allows the physical environment to absorb changes that would otherwise require corresponding changes in human spatial knowledge.
If a resource moves, the request need not change. The resolution function does.
5Minimum Sufficient Intelligence
An important consequence follows from this formulation. An intent-addressable physical environment does not necessarily require maximal artificial intelligence. It requires sufficient intelligence to produce a reliable physical resolution. This distinction is important within a technology landscape increasingly oriented toward large foundation models and generalized reasoning systems. Edge-intelligence research has demonstrated both the value and the difficulty of moving inference toward resource-constrained environments, where latency, energy, memory, privacy, reliability, and connectivity become fundamental architectural considerations [4][5].
For many physical resource-resolution tasks, the objective is considerably narrower than unconstrained natural-language reasoning. The system may need to determine which of a bounded set of physical resources best satisfies an intent under a bounded set of contextual conditions, then produce a correspondingly bounded physical response.
A physical-resolution system should employ the minimum intelligence required to resolve the relevant intent reliably under defined operating conditions.
Principle of minimum sufficient intelligence
The objective is not to minimize capability arbitrarily. Nor is it to maximize model complexity. The objective is to minimize unnecessary computational complexity while satisfying the operational requirements of the physical task. Three consequences follow.
- System intelligence becomes subordinate to physical performance. A model is useful to the extent that it improves correct, timely, and dependable physical resolution.
- System design can explicitly prioritize bounded behavior. Where the action space and resource population are constrained, the resolution process can be correspondingly constrained.
- Local inference becomes more practical. Edge intelligence has particular relevance where response latency, resilience, bandwidth, privacy, and network independence matter [4][5].
In this context, Physical AI need not always mean placing the largest available model closest to the physical world. For some classes of infrastructure, the appropriate engineering objective may instead be the smallest reliable intelligence capable of closing the physical loop.
6Search as an Infrastructure Function
The practical implication of IAPI is a change in the allocation of work between people and environments. In conventional resource-dense environments, physical search frequently requires humans to maintain an internal model of the environment. They learn storage conventions. They remember locations. They read labels. They scan shelves. They backtrack. They ask other people. These activities are so familiar that they are rarely treated as computation. But functionally, they constitute a resolution process. The human receives an intent, generates candidate locations, evaluates observations, updates hypotheses, and terminates the search when the desired resource is found.
IAPI asks whether that resolution process should remain a human responsibility when the environment itself can perform it. In a hospital, the difference could be between a clinician searching multiple supply locations and an environment identifying the relevant stocked location directly. In manufacturing, it could be the difference between a technician recalling where a specialized tool was returned and infrastructure resolving its current endpoint. In logistics, it could be the difference between workers navigating storage topology manually and infrastructure translating requested resources directly into physical guidance.
These are different operational environments, but the architectural primitive is the same. The person or machine specifies the what. The infrastructure determines the where.
7IAPI and Existing Physical Resource Technologies
Real-time location systems (RTLS) can establish or estimate where a tracked asset is. Pick-to-light systems can indicate a location selected by an upstream workflow. Voice-directed systems can provide an interaction modality for executing structured work. Each can contribute important capabilities to an intent-addressable environment, but none by itself defines the IAPI abstraction.
| Technology | Primary function |
|---|---|
| RTLS | Establish or estimate where a tracked asset is. |
| Pick-to-light | Indicate a location selected by an upstream workflow. |
| Voice-directed systems | Provide an interaction modality for workflow execution. |
| IAPI | Resolve what is needed to the physical endpoint that should satisfy the request now. |
The distinction is architectural rather than technological. IAPI defines the responsibility for request-time resolution: the requester specifies the required resource or outcome without supplying its physical locator; infrastructure resolves that intent against current physical state to an appropriate physical endpoint and produces guidance or actuation. Accordingly, RTLS, pick-to-light, voice interfaces, computer vision, inventory systems, or robotic actuation may operate beneath or within an IAPI implementation. IAPI does not require any one of them. Its defining function is the mapping from expressed intent to an appropriate physical resource endpoint under current conditions.
8Implications for Physical AI
If Physical AI continues to move intelligence into the built environment, systems will require more than perception models and actuators. They will also require abstractions governing how humans and machines refer to, resolve, and interact with physical resources. IAPI is one proposed abstraction for that layer.
- Intent is separable from location. A requester need not know the physical locator of the requested resource.
- Physical resources are resolvable endpoints. Infrastructure maintains sufficient state to associate a request with an appropriate physical resource.
- Resolution occurs at request time. The infrastructure determines the relevant endpoint under current conditions rather than requiring location to be permanently encoded in the request.
- Resolution terminates in the physical world. The output is not merely informational. It produces guidance, indication, routing, or physical actuation.
- Intelligence is bounded by the physical task. The computational architecture is evaluated by whether it reliably performs the required resource resolution, not by model complexity alone.
Together, these properties distinguish IAPI from a generic intelligent environment. The central function is not simply that the environment can compute. It is that the environment can resolve intent into physical addressability.
9Conclusion
Physical AI represents a broader transition from computational systems that principally model information toward systems that perceive, reason about, and act within the physical world. That transition creates a corresponding need for infrastructure abstractions that determine how physical resources are requested and resolved. Intent-Addressable Physical Infrastructure defines one such abstraction.
Humans or software agents should be able to specify what they need without necessarily knowing where it is.
The central proposition
The infrastructure assumes responsibility for resolving that intent against the current physical environment and producing the appropriate guidance or actuation. Under this model, physical search is no longer necessarily an intrinsic human task. It becomes a function that can be assigned to intelligent infrastructure. If Physical AI concerns the growing ability of computational systems to operate intelligently within the physical world, IAPI concerns a more specific question: how should that physical world become addressable? The answer proposed here is: through intent.
Implementation note. Ambi, developed by Intelligent Clinical Systems, is one implementation of IAPI for physical resource resolution in clinical supply environments.
- [1]E. R. Griffor, C. Greer, D. A. Wollman, M. J. Burns. Framework for Cyber-Physical Systems: Volume 1, Overview. NIST SP 1500-201, 2017.
- [2]From embodied intelligence to physical AI. Nature Machine Intelligence 8, 491-492, 2026.
- [3]L. Zhang, A. Afanasyev, J. Burke, V. Jacobson, K. C. Claffy, P. Crowley, C. Papadopoulos, L. Wang, B. Zhang. Named Data Networking. ACM SIGCOMM Comput. Commun. Rev. 44(3), 2014.
- [4]Z. Zhou, X. Chen, E. Li, L. Zeng, K. Luo, J. Zhang. Edge Intelligence: Paving the Last Mile of Artificial Intelligence With Edge Computing. Proc. IEEE 107(8), 2019.
- [5]S. Deng, H. Zhao, W. Fang, J. Yin, S. Dustdar, A. Y. Zomaya. Edge Intelligence: The Confluence of Edge Computing and Artificial Intelligence. IEEE Internet Things J. 7(8), 2020.