Intent-Bound Elder Care and Companion Robotics

Regulatory and Domain Context

Elder-care robotics has moved from research demonstration to market reality. Intuition Robotics' ElliQ has been distributed at scale by state agencies on aging in New York and Florida. SoftBank's Pepper and Sony's aibo have been deployed in Japanese long-term-care facilities for over a decade, and a growing class of socially-assistive robots — Mabu, Stevie, Catalia Health's offerings, and a long line of MIT Media Lab spinouts — operate in private homes, assisted-living settings, and skilled-nursing facilities. Toileting, transfer-assist, and mobility platforms from Toyota, Panasonic, and Cyberdyne are entering U.S. clinical pilots. Each platform performs actions whose appropriateness depends not only on what is mechanically safe but on who has the authority to direct that action toward this resident at this moment.

The regulatory environment around these deployments is consolidating quickly. The FDA has clarified that socially-assistive and physically-assistive robots performing functions tied to clinical outcomes are medical devices, with classification depending on the intended use and the level of clinical decision-making the device performs or supports. ISO 13482 establishes safety requirements for personal care robots and explicitly contemplates the relationship between the robot, its operator, and the cared-for person. CMS reimbursement frameworks for remote patient monitoring, chronic care management, and emerging technology-assisted-care codes increasingly require documentation of who authorized a given intervention and on what basis. State law on guardianship, supported decision-making, and POLST/MOLST orders defines layered, sometimes conflicting authority structures over the daily life of an elderly person, and dignity-preserving authority delegation — the principle that authority should be exercised at the least restrictive level consistent with safety — is becoming an explicit design expectation rather than an aspirational ethics statement.

Architectural Requirement

The architectural problem this regulatory environment exposes is not a single-authority problem. The resident retains primary authority over their own person, expressed through ongoing consent, advance directives, and supported decision-making arrangements. The professional caregiver — nurse, aide, therapist — holds clinical authority bounded by license, scope of practice, and the resident's care plan. The family member or designated proxy holds delegated authority that is itself bounded by documents, statutes, and the resident's capacity. The institution holds authority over policy, scheduling, safety protocols, and the platform's configuration envelope. The physician of record and the payer hold authority over interventions that touch clinical or reimbursable domains. A correct robot action is one that lies inside the intersection of every applicable authority at the moment of action, and a correct audit record is one that can show the authority composition that admitted the action.

Authority over a care robot is also temporally fluid in ways unique to this domain. A resident's capacity may fluctuate within a single day. A caregiver's shift hands off every eight or twelve hours. A family proxy's authority may activate or lapse with a clinical event. The institution's policy envelope shifts with regulatory updates and survey findings. The platform must compose authority at every action while tracking its evolution, and it must distinguish between actions that require fresh, high-fidelity authority — medication prompts, transfers, contact with the resident's body — and actions that can proceed under standing, lower-fidelity intent, such as ambient companionship or environmental monitoring. Graduated fidelity is essential: requiring a full multi-authority composite for every micro-interaction is operationally impossible, while admitting clinically significant actions under ambient defaults is unsafe and non-compliant.

Why Procedural Compliance Fails

The dominant approach in deployed elder-care robotics is procedural: a consent form at intake, a care plan in the EHR, a facility policy binder, and an after-the-fact incident report when something goes wrong. Procedural compliance fails for three reasons specific to this domain. First, it is silent at the moment of action. A robot deciding whether to administer a medication prompt, initiate a physical assist, or escalate a perceived emergency cannot consult a paper consent form; it can only consult a runtime structure that refuses unauthorized behavior. Second, it cannot represent dignity-preserving delegation. The principle that authority should be exercised at the least restrictive level — that a resident's own preference governs unless and until capacity or safety considerations require escalation — is a compositional rule that procedural artifacts cannot encode. Third, procedural records are not the structural trace that FDA post-market surveillance, CMS audit, and institutional risk management increasingly require. A signed care plan does not show which authority admitted the specific action that caused the adverse event.

The dignity costs of procedural failure are not abstract. Robots that override resident preference because the institutional default was set conservatively, robots that defer to a family member when the resident retains decisional capacity, robots that proceed with a clinical action because no caregiver was present to refuse it — each is a structural failure misdescribed as a procedural one.

What the AQ Primitive Provides

Operator-intent treats the elder-care authority stack as a runtime structure. The resident's intent is the innermost and most strongly weighted layer, expressed through credentialed consent channels and, where capacity is established, sufficient on its own to admit ordinary daily-living interactions. The caregiver-of-record's intent is a medium-fidelity layer, refreshed at handoff and gated by a credential tied to license and assignment. The family-proxy intent activates only under documented conditions and within explicit scope. The institutional-policy intent supplies the outermost envelope: prohibited interventions, mandatory escalations, scheduling and configuration constraints expressed as machine-evaluable rules rather than binder text. Physician and payer intents enter the composition for actions that cross into clinical or reimbursable domains.

Every behavior the robot considers is admitted through a composite check across the layers applicable to that behavior's fidelity tier. Ambient companionship admits under standing resident intent and the institutional envelope. Medication prompts admit only when a current caregiver-of-record intent and an active clinical order are both present. Physical-assist actions require a fresh caregiver attestation and a resident consent that has not been withdrawn. Escalations to family or clinical staff are themselves admitted actions, with their own intent composition and audit trace. Multi-fleet and multi-facility operations — a chain of skilled-nursing facilities, a home-care agency operating across institutional boundaries — compose authority through the same primitive, so that a robot moving from one care environment to another carries its authority structure with it rather than being reconfigured by hand.

Crucially, the primitive supports dignity-preserving delegation by construction. Authority composition begins at the resident layer and escalates only when capacity, safety, or scope demands it. The audit trace shows not only which authority admitted each action but which authorities were available and not invoked, making least-restrictive practice visible rather than merely asserted.

Compliance Mapping

The mapping to current and emerging regulation is direct. FDA medical-device requirements for software in medical devices and for human-factors validation map onto the credentialed-intent channels and the structural audit trace, which together demonstrate that each clinically significant action was authorized by an identified, credentialed party. ISO 13482's safety requirements for personal care robots and its expectations around operator and user roles map onto the layered intent structure and the explicit fidelity tiers. CMS documentation requirements for reimbursable interventions map onto the per-action audit record showing which clinical authority admitted the intervention. State guardianship and supported-decision-making law map onto the resident-primary, escalating-on-condition composition the primitive enforces. EU AI Act provisions for high-risk AI in healthcare and assistive contexts map onto the same logging and human-oversight infrastructure that operator-intent makes mandatory. Dignity-preserving authority delegation is not a separate compliance regime but the architectural default the primitive expresses.

Adoption Pathway

Adoption begins where the operational and regulatory pressure is greatest: skilled-nursing facilities operating physical-assist or medication-prompt platforms under active CMS and state survey scrutiny. The first deployment stage replaces ad hoc consent forms and verbal handoffs with credentialed resident and caregiver intent channels, producing an immediate audit improvement and a defensible posture against survey findings. The second stage incorporates institutional-policy intent as a machine-evaluable envelope, allowing facilities to update protocols centrally and have those updates propagate to every platform without firmware revisions. The third stage brings family-proxy and physician-of-record intent online for clinically adjacent actions, supporting CMS-reimbursable workflows and FDA post-market expectations. By the time a deployment expands to home-care and multi-facility operation, the primitive already carries the authority structure those settings require, and the dignity-preserving posture that elder-care robotics is increasingly required to demonstrate is provided by construction rather than by policy assertion.