Vendor and Product Reality
1X Technologies, formerly Halodi Robotics, is led by Bernt Bornich, is headquartered in Moss, Norway, and maintains engineering operations in the United States. The company shipped EVE, a wheeled humanoid torso, into industrial security and logistics pilots, and used those deployments to build the manipulation and tele-operation stack that NEO inherits. NEO Beta was unveiled in 2024 with an explicit consumer-home target, soft outer shells, tendon-driven actuators, and a stated design intent of operating safely around humans without the rigid-body force profile of competitors like Figure or Tesla Optimus.
The OpenAI relationship is both financial and architectural. OpenAI's Startup Fund is among 1X's investors, and the companies have publicly described joint interest in language-conditioned manipulation and embodied reasoning. NEO is positioned as a platform on which OpenAI-class models drive low-level control through a learned policy rather than through scripted behavior trees, and the consumer roadmap depends on that learned stack generalizing across the long tail of household tasks.
The product story 1X tells publicly is one of in-home assistance: laundry, tidying, fetching, basic eldercare presence. That story implies multiple humanoids per home over time, humanoids cohabiting with non-humanoid robots like vacuums and lawnmowers, and humanoids visiting from service providers. The fleet topology is mixed by construction, and the operator, the human in the home, is non-technical, non-credentialed in any robotics sense, and emotionally invested in outcomes the robot may not understand.
Architectural Gap
NEO's control stack assumes a single tight loop between the household operator's natural-language instruction and the on-board policy that executes it. That loop works for unambiguous tasks in a single-robot home, but it breaks in two directions as the deployment matures. Outward, when a second humanoid or a non-1X robot enters the home, there is no architectural object that represents the operator's intent at a fidelity multiple agents can consume; each robot resolves the instruction independently against its own model and its own world view. Inward, when the operator's intent is partial, "tidy up before my mother arrives", the policy must guess at fidelity, and it has no graduated channel for asking the operator to refine.
The mixed-fleet case is where the gap shows first commercially. A household running a NEO, a Roomba j7+, a Husqvarna Automower, and a delivery robot from a grocery service has four agents whose work composes, the NEO does not vacuum where the Roomba is currently running, the Automower does not block the delivery robot's arrival path, and today that composition is achieved either through brittle pairwise integrations or not at all. None of the agents can read the operator's intent at a shared fidelity.
The safety case is where the gap shows first for auditability. As consumer-AI and product-safety regimes mature, they trend toward requiring that operator intent in safety-relevant actions be auditable and attributable to an authorizing human. A single-loop stack records what was instructed and what was done; it does not, on its own, record which authority graded which fidelity tier and what the operator was given the chance to refine before a safety-relevant action executed.
What the Operator-Intent Primitive Provides
The operator-intent primitive supplies graduated fidelity tiers for mixed-fleet coordination. Operator intent is modeled as a structured object with explicit fidelity levels: coarse outcome ("home is presentable"), task decomposition ("laundry done, dishes put away, floor clear"), action constraints ("do not enter the nursery before 7am"), and operator-acknowledged risk acceptance ("you may move the espresso machine"). Each tier is signed by the operator at the moment of declaration, and the agents executing against the intent declare which tier they are operating at and what they have escalated for refinement.
Multi-fleet intent fusion is the operational core of the primitive. The same operator intent object is consumed by every agent in the home, NEO, the vacuum, the lawnmower, the visiting delivery robot, and each agent contributes its planned actions back to the object so that conflicts surface architecturally rather than through physical near-misses. The fusion engine resolves authority precedence (the operator's nursery-exclusion overrides any agent's tidying plan), surfaces fidelity gaps (no agent has claimed responsibility for the espresso machine), and produces an attestable record of what the operator authorized and what the fleet did with that authorization.
The graduation matters because consumer operators do not specify intent at uniform fidelity. The primitive lets the operator stay coarse where they want to and forces refinement only where safety, multi-agent conflict, or regulatory audit demands it. The operator's experience is closer to delegating to a household manager than to programming a robot, and the regulatory record is closer to a signed instruction than to a log file.
Composition Pathway
A pragmatic adoption path begins inside the NEO control stack. 1X exposes the on-robot intent object as a first-class artifact rather than as an internal interpretation of a natural-language string. The OpenAI-class model that today renders an instruction into a policy plan instead renders it into a graduated intent object, signs that object on behalf of the operator after acknowledgment, and runs the policy against the object. NEO's behavior does not change for simple instructions; it gains structure for compound and safety-relevant ones.
Multi-NEO homes compose next. As 1X ships second and third units into the same household, the intent object becomes the shared coordination surface, and the units arbitrate through declared fidelity rather than through ad-hoc co-presence logic. This is where the substrate first earns its keep without external partners, 1X owns both endpoints and can demonstrate the architecture cleanly.
Cross-vendor mixed fleet is the third layer. iRobot's Roomba, Husqvarna's Automower, Amazon's Astro, and grocery-delivery robots all subscribe to the same operator-intent object and contribute their planned actions back. The integration contract is the intent schema, not a bilateral API per vendor pair, and 1X, having shipped the substrate first, becomes the natural reference implementation. The household operator interacts with one intent surface; the robots beneath it coordinate without operator involvement.
Commercial Implication
The consumer-humanoid commercial story today is unit sales: 1X sells NEO units at a hardware-plus-subscription price, and revenue scales with units in homes. The operator-intent substrate adds a second surface: 1X licenses the intent platform to peer robot vendors who need to participate in NEO households, and to insurers, regulators, and home-services providers who need to read or write intent on the operator's behalf. That platform revenue is uncorrelated with NEO unit sales and scales with the household graph rather than with the 1X manufacturing line.
Insurance is the most concrete near-term participant. Home insurers who today cannot underwrite humanoid-related liability gain a signed operator-intent record that disambiguates which actions the operator authorized and which the agent escalated. The insurer becomes a co-reader of the intent object and a counterparty to 1X's safety claims, and the resulting policy is materially cheaper for the household. That price difference is itself a sales driver for NEO against humanoid competitors that ship without the substrate.
The defensive case is about who authors the reference. Product-safety and consumer-AI expectations trend toward credentialed, graduated operator authority for safety-relevant household actions, and humanoid platforms will be among the first products examined against those expectations. A vendor that ships credentialed operator intent first is positioned to define the working reference for compliance; a vendor that ships later inherits a schema shaped by whoever moved first, whether that is a peer such as Figure, Tesla Optimus, or Apptronik Apollo, or a neutral third party.
Licensing Implication
Operator-intent is an Adaptive Query architectural primitive, and the substrate's value depends on the intent object being identical across vendors. A NEO and a Roomba and an Astro that each interpret a coarse household instruction through divergent intent schemas produce no composition; they produce three uncoordinated agents in the same room. A licensed implementation guarantees the schema is shared; parallel reimplementations guarantee it is not. For a consumer-humanoid vendor whose product depends on cohabiting with non-1X robots, the license is the mechanism that makes cohabitation work at all.
Taken early, the license also gives 1X first authorial position over the fidelity tiers, the escalation semantics, and the operator-acknowledgment ceremony that the rest of the consumer-robot industry will eventually conform to. Taken late, the license is still available, but the schema reflects the priorities of whoever moved first. Given 1X's OpenAI alignment and stated consumer-AI safety posture, the architectural opportunity to author that schema is unusually well-matched to the company's existing positioning, and the cost of ceding it to a peer is correspondingly high.
How to Build It
A skilled implementer can realize the operator-intent substrate on a NEO-class platform without changing the learned control policy. The build has four cooperating parts.
First, an intent object. Model operator intent as a signed, structured record carrying an identifier, the authorizing operator's credential, an ordered set of fidelity tiers (from coarse outcome, through task decomposition, through explicit action constraints, through operator-acknowledged risk acceptance), an expiry, and a revocation handle. The credential can be realized with any standard asymmetric-signature or verifiable-credential scheme; the object is bounded (it names what is authorized and what is excluded) and revocable (the operator or an authority can invalidate it, after which downstream actuation against it must stop).
Second, an intent envelope check at the actuation boundary. Before a policy plan drives actuators, the plan is checked against the tier and constraints of the intent it cites. Actions inside the envelope proceed; actions that would exceed it are deferred, and the agent escalates to the operator for refinement or a higher-fidelity authorization rather than guessing. This gate is where meaningful human control becomes structural rather than advisory.
Third, a fusion surface for mixed fleets. Each agent in the home, humanoid or not, subscribes to the same intent object and writes back its planned actions. A resolver applies authority precedence (an operator exclusion overrides any agent plan), surfaces fidelity gaps where no agent has claimed a task, and detects composition conflicts before they become physical near-misses. The integration contract is the shared intent schema, not a bilateral API per vendor pair.
Fourth, a lineage record. Every governed action is bound to the intent object and the operator credential that authorized it, at the fidelity tier under which it executed, producing an attestable trail from action back to authorizing human.
Embodiments and variations. The operator credential can be a household account, a per-person credential, or a delegated service-provider credential. Fidelity tiers can be as few as two (coarse outcome plus explicit exclusions) or many; tiers can be declared ahead of time or negotiated interactively at escalation. The envelope check can run on-robot, on a home hub, or split across both. Intent can be authored by voice, app, or a caregiver on the operator's behalf under a delegation credential. The fleet can be all-1X (multiple NEO units), 1X plus consumer robots (a robot vacuum, a robotic mower, a home monitor), or 1X plus visiting service robots. The substrate is not specific to humanoids or to the household; the same intent object, envelope check, fusion surface, and lineage record apply to any setting where an operator delegates bounded authority to one or more autonomous units that must defer or escalate at the edge of that authority.
Disclosure Scope
The invention described here, the Operator Intent inventive step, is disclosed in U.S. Provisional Application No. 64/049,409. The claims about what the operator-intent substrate does (credentialed, bounded, revocable intent objects; graduated fidelity tiers; envelope-bounded actuation with defer-and-escalate behavior; multi-agent intent fusion; and action-to-operator lineage) trace to that filing.
References to 1X Technologies, NEO, EVE, OpenAI, and other named products and companies, and any statements about the consumer-robotics market, regulation, or insurance, are external context provided to situate the disclosure against the current state of the art. They are described as public fact and are not claims of the provisional application. Product names are the marks of their respective owners and are used here for identification and comparison only.
