NSF_X-Labs_Physical_AI_Out-of-Scope_Annex.pdf

NSF X‑Labs AI for Physical Systems Annex – Out‑of‑Scope

Purpose: Clarifies topics that are likely out-of-scope and nonresponsive to the Artificial Intelligence (AI) for Physical Systems Topic and provides a quick screening checklist for applicants.

The examples provided in this Annex are illustrative and do not establish additional evaluation criteria. Proposals will be evaluated using the selection criteria stated in Sections 11 and 13 of the NSF-OTASO-FY26- XLabsInitiative or latest amendment.

Physical AI definition: Intelligent, adaptive, scalable AI systems embodied in or interfacing with the physical world, which may include (but not be limited to): robots and cyber-physical systems; implantable, ingestible, or wearable AI devices; neural or bio-integrated platforms; and built-environment/infrastructure-scale systems.

Out‑of‑scope examples

• Mission fit & platform scope: Narrow, single product efforts that do not generalize into reusable platform technology across domains; integration “glue” around existing tech that yields only incremental improvements; fundamental theory or algorithms without hardware or wetware-in-the-loop validation;

proposals primarily focus on testbeds, data hubs, or cloud infrastructure as the mission rather than enablers within a platform advancing Physical AI.

• Technical content (what the platform is/does): Simulation- or digital-twin-only work lacking plans to demonstrate sim-to-real transfer; large language model (LLM) wrappers or policy tweaks that control commercial off-the-shelf robots without new embodied capabilities in perception-action-learning; narrow device-layer add-ons, instrumentation, User Interface (UI) or dashboards, teleoperation apps, single-sensor improvements, or generic wearables that do not produce closed-loop autonomy/adaptability; biological cataloging or data-lake plays that do not yield novel embodied control or controlled training datasets tied to platform validation; speculative bio-AI (e.g., unconstrained synthetic neural implants) without feasible in-vitro/in-vivo validation, safety mitigations, and regulatory and ethical oversight; single-indication medical devices primarily seeking trial or market scale-up funding without overcoming a fundamental embodied-AI barrier; services or tooling-only efforts that do not deliver predictable, controllable embodied behavior in biological/physical systems.

• Maturity boundaries: Not aligned to early-stage platform development [~Technology Readiness Levels (TRLs) 2–4] with clear milestones to mature via integrated design-build-test-learn; concepts unlikely to reach credible platform readiness (e.g., hardware or wetware-in-the-loop demonstration and external pilot users) within ~5–7 years; or technologies already near market that primarily need private capital for scale-up without major technical barriers remaining.

• Safety, responsible autonomy, and compliance: Human-robot interaction or embodied AI in or around humans that do not consider issues such as security, failure-mode analysis, benchtop/sandbox testing, independent oversight, reliability considerations, etc.; uncontrolled field or clinical deployments of autonomous or bio-integrated systems without appropriate approvals, layered containment and controls, and third-party review [e.g., Institutional Review Board (IRB); Institutional Animal Care and Use Committee (IACUC); other medical device pathways].

• Redundancy with existing U.S. investments: Advancement or scaling of technologies already supported by substantial public or private R&D unless the proposal clearly introduces a new platform capability not covered by those investments.

• Nontechnical or pure commercialization-only efforts: Missions where the challenge is solely business development, market access, manufacturing scale‑up, etc., with no new technical platform work or adding a platform capability.