Figure Humanoid Robotics

Figure Reality

Figure AI operates Figure 02, the second-generation bipedal humanoid platform that succeeded the Figure 01 prototype, with active manufacturing deployment at BMW's Spartanburg, South Carolina plant. The BMW engagement — initially a sheet-metal handling and chassis-staging pilot, expanding into broader cell-level industrial work — represents the most visible production humanoid deployment among the current cohort of Western humanoid programs. Figure's commercial trajectory rests on the Helix vision-language-action model, an end-to-end neural policy that maps natural-language operator instruction and onboard vision directly into bimanual manipulation behavior, replacing the discrete skill-library architectures common in earlier humanoid programs.

The corporate context is decisive for trajectory. Figure's funding cohort — Microsoft, OpenAI, NVIDIA, Jeff Bezos, Intel Capital, Parkway Venture Capital — and the OpenAI collaboration that produced the conversational-agent demonstrations on Figure 01 placed the program at the center of the foundation-model-meets-embodiment narrative. The subsequent in-housing of the language-and-behavior stack into Helix, and the announced separation from external foundation-model dependencies for the on-robot policy, repositions Figure as a vertically-integrated humanoid operator rather than a hardware shell wrapped around someone else's model.

What Figure has demonstrated is a humanoid platform capable of executing an industrially-meaningful workload under a single operator's instruction at a single site under a single OEM customer's authority. What no humanoid program has yet demonstrated — and what the industrial-deployment trajectory will inevitably demand — is multi-fleet, multi-authority operator-intent composition.

Emerging Industrial Deployment

Industrial-deployment humanoid operations face emerging worker-safety oversight and emerging facility-operator authority composition requirements that have no analog in the controlled-pilot deployments visible today. OSHA general-duty oversight, robotics-specific ANSI/RIA R15.06 and ISO 10218/15066 collaborative-robot standards, and emerging humanoid-specific guidance from NIST and the EU AI Act high-risk-system framework all converge on a common architectural premise: the operator whose intent governs a humanoid's behavior must be identifiable, credentialed, and accountable, and that credential must compose across the multiple authorities — facility operator, OEM customer, fleet operator, integrator, regulator — that share governance over a deployed humanoid.

BMW Spartanburg is one site under one OEM authority. The trajectory is multi-OEM, multi-site, multi-shift, and multi-fleet — Figure 02 units operating across BMW, a separate automotive customer, a logistics 3PL, and an emerging consumer-facing deployment, each under a different facility-operator authority and a different regulatory regime. Helix as a behavior policy is silent on this composition; it accepts an instruction and produces a behavior. The architectural element that makes that instruction credentialed, that makes the operator identifiable, and that composes the resulting authority across fleets is operator-intent.

Operator-Intent Substrate

Operator-intent is the architectural primitive for graduated-fidelity, multi-fleet, multi-authority intent. Operator intent — facility manager, OEM customer, shift supervisor, integrator, regulator — admits through credentialed declarations rather than implicit instruction. Each authority emits a signed intent artifact at the fidelity tier appropriate to its role: a facility manager might declare a coarse work-envelope and safety geometry; an OEM customer might declare a task-level objective; a shift supervisor might declare a per-unit dispatch instruction; a regulator might declare a compliance constraint that gates the entire envelope.

Graduated fidelity tiers are the load-bearing element. A coarse envelope from a facility manager composes with a task objective from an OEM customer composes with a dispatch instruction from a shift supervisor composes with a compliance constraint from a regulator — each at its native fidelity, each bound to its issuing authority's credential, each composing into a single composite admissibility decision that gates Helix's behavior policy at execution time. Multi-fleet composition follows the same pattern: a Figure 02 unit operating under simultaneous instruction from a BMW shift supervisor and a Figure fleet operator resolves the composite intent through the substrate rather than through ad-hoc precedence rules baked into the controller.

Why The Layer Is Architectural

The temptation in early humanoid deployment is to treat operator authority as a configuration concern — a per-site policy file, an integrator's runbook, an OEM-customer's standing instruction. That treatment scales to one site under one authority. It does not scale to the multi-OEM, multi-fleet, multi-regulator trajectory that the Figure roadmap, the Helix model, and the broader humanoid-industrial sector are visibly heading toward. Once a single Figure 02 unit operates under simultaneous and partially-conflicting instruction from a facility manager, an OEM customer, a fleet operator, and a regulator, the composition problem is no longer a configuration concern; it is an architectural primitive.

Operator-intent provides that primitive. The substrate sits above Helix — Helix continues to map instruction-and-vision into behavior — and below the multi-authority instruction surface. Composite admissibility supports the multi-authority operations that emerging industrial deployment requires; graduated fidelity tiers support the heterogeneous-authority composition that any cross-customer humanoid fleet will encounter; credentialed declarations support the worker-safety and regulatory oversight that humanoid deployment will not be permitted to operate without.

Figure Position

Figure gains architectural substrate for emerging industrial humanoid deployment without modifying Helix, without renegotiating the BMW engagement, and without absorbing multi-authority composition into the on-robot policy stack. The platform's commercial trajectory — from single-site pilot to multi-OEM, multi-fleet industrial operator — depends on exactly this kind of architectural element above the behavior policy. Operator-intent is the substrate that lets Figure 02 operate credentialedly across the fleets, sites, and authorities the next phase of humanoid industrialization will demand.

The competitive context — Tesla Optimus, Agility Digit, Apptronik Apollo, 1X Neo, Boston Dynamics Atlas, Unitree H1 — is converging on a similar industrial-deployment thesis under similar regulatory pressure. The vendors who treat operator authority as a configuration concern will encounter the multi-authority composition problem at the same point in their trajectory; the vendors who adopt an operator-intent substrate above their behavior policy will absorb worker-safety oversight, OEM-customer authority, and fleet-operator dispatch as composable layers rather than as ad-hoc precedence rules. Figure's vertical-integration posture around Helix makes it structurally well-positioned to adopt operator-intent as the architectural element above the policy stack, preserving Helix's end-to-end advantage while gaining the credentialed multi-authority composition that no end-to-end policy alone can provide.