What can humanoid robots actually do today?

The capability framework: four verification tiers

What humanoid robots actually do in 2026 sorts into four structurally distinct verification tiers per DEPLOY's framework on capability claims. Each tier carries a different verification posture:

• Verified consumer-deployed: capability that ships to consumer customers, runs in customer homes, and produces verifiable consumer-deployment outcomes. Only one cohort manufacturer operates at this tier: 1X Technologies NEO.
• Verified enterprise-deployed: capability that operates in customer facilities under enterprise contracts, with verifiable per-deployment evidence. Multiple cohort manufacturers operate at this tier: Figure AI, Apptronik, Agility Robotics, and UBTech.
• Research and demonstration: capability documented in research papers, marketing demonstrations, and event footage, but not consumer-deployed or enterprise-deployed at scale. Most cohort manufacturers operate research and demonstration alongside their deployed work: Boston Dynamics Atlas, Tesla Optimus, and Unitree G1 + R1 sit primarily at this tier.
• Claimed future: capability promised in maker communications and product roadmaps, not currently demonstrated at verifiable depth. Cooking, autonomous home assistance, childcare, full-spectrum household work, and (for several makers) Tesla-Optimus-style consumer general-purpose use sit here.

The framework partitions capability claims into these four tiers because operators and consumers reading humanoid marketing content benefit from knowing which tier any specific claim actually anchors against.

Tier 1: Verified consumer-deployed

1X NEO is the only consumer-deployed humanoid in 2026. NEO performs:

• Laundry tasks (folding, sorting, light loading) in customer homes. See can humanoid robots do laundry.
• Light manipulation tasks (object placement, organizing, fetch-and-carry).
• Limited household assistance under Expert Mode teleoperation for complex tasks.
• Companionship and presence in domestic environments.

The verification posture is verified-by-disclosure: 1X publishes teleop reliance explicitly on the consumer commerce surface, with Expert Mode framing that allows customer-controlled operator sessions for complex tasks. Operators evaluating NEO know what they are buying.

Tier 2: Verified enterprise-deployed

Four cohort manufacturers operate verified enterprise deployments:

• Figure AI: Figure 02 at BMW Spartanburg (30,000 X3 vehicles, 11-month deployment, end-product OEM acceptance). Figure 03 at Catalyst Brands Reno (pilot). Manufacturing and logistics task scope.
• Apptronik Apollo: Mercedes-Benz Berlin-Marienfelde + Hungarian assembly; GXO Logistics multi-phase R&D-to-distribution; Jabil manufacturing partnership. Three Fortune-500 enterprise pilots.
• Agility Robotics Digit: GXO Flowery Branch with 100,000-tote throughput under multi-year RaaS contract. Single-customer commercial depth.
• UBTech Walker S2: BYD, Geely, Foxconn factory pilots in Chinese industrial ecosystem. HKEX-listed public company.

Verified enterprise deployment is materially different from consumer deployment. Enterprise customers integrate the robot into manufacturing or logistics workflows under contractual scope; the deployment envelope is controlled, the tasks are defined, and the human-in-loop posture is operationally established. Consumer customers cannot buy these robots; per-unit pricing is enterprise-contract-bound.

Tier 3: Research and demonstration

The research and demonstration tier holds substantial capability but does not deploy at consumer or enterprise scale:

• Boston Dynamics Atlas: research engineering platform with advanced dynamic-mobility and manipulation demonstrations. The new electric Atlas (April 2024 reveal) is positioned for commercial transition; commercial deployment is pending.
• Tesla Optimus: factory-internal pilots inside Tesla's own facilities (which DEPLOY's framework classifies as research per the maker-facility rule); marketing demonstrations of folding clothes, serving drinks, and basic manipulation. No third-party customer deploys Optimus.
• Unitree G1 + R1: research-tools positioning. The platforms ship to research and developer customers (not consumer-grade deployment); the capability scope is research-environment manipulation and locomotion, not commercial-application work.

The research and demonstration tier is editorially substantive. Engineering credibility, research output, and capability progress all happen here. The framework does not treat research as inferior to deployment; it treats them as distinct verification surfaces.

Tier 4: Claimed future

The claimed-future tier collects capability that is promised in marketing communications but not yet demonstrated at verifiable depth:

• Cooking (see can humanoid robots cook): claimed across multiple makers; not consumer-deployed or enterprise-deployed.
• Full-spectrum household work: claimed in Tesla Optimus's consumer-promised tier; not currently consumer-deployed.
• Childcare: claimed in some maker communications; not consumer-deployed; raises substantive liability and safety questions that have not been addressed at commercial-deployment depth.
• Autonomous home assistance (full-spectrum): claimed across the cohort; the NEO consumer-deployed work is a subset of the broader claim, not the full claim.
• Driving a car: ambiguous claim conflating Tesla Optimus (humanoid; not designed to drive) with Tesla Robotaxi (autonomous vehicle service). See Tesla Robotaxi vs Tesla Optimus.

The framework reads claimed-future capability as editorially significant claim layer, not as gap. Maker communications about future capability shape operator expectations + investor positioning + customer marketing; verification asks where the claim sits relative to current demonstrated work.

Why the framework matters

Operators evaluating "what can humanoid robots do" benefit from knowing which tier any specific claim sits in. A marketing demonstration of a humanoid folding clothes is research and demonstration tier; the same task verified at consumer deployment (1X NEO) is tier 1. Both might involve clothes-folding; they are at different verification states.

Per DEPLOY's vvc-sharper-across-competitive-set discipline, the four-tier framework operates alongside the five-tier availability framework (which addresses "can you buy it") to produce a two-dimensional reading: availability tier + capability tier. A consumer-available + verified-consumer-deployed humanoid (NEO at consumer tasks within its scope) is a different artifact than a consumer-promised + claimed-future humanoid (Tesla Optimus at general-purpose household work).

Where to go for context

For per-maker capability detail, see the cohort entity explainers: 1X NEO, Figure 03, Apptronik Apollo, Boston Dynamics Atlas, Tesla Optimus, UBTech Walker S2, Unitree G1 + R1, Mentee Robotics, Sanctuary AI Phoenix.

For task-specific capability evaluation, see can a Tesla Optimus clean a house, can humanoid robots cook, can humanoid robots do laundry.

For the complementary availability framework, see can I buy a humanoid robot in 2026. For the canonical DEPLOY verification framework across capability and deployment, see how DEPLOY verifies capability claims and what verified means at DEPLOY. For methodology canonical references applicable to current-capability framing: the 4-way autonomy-boundary taxonomy + verified-vs-claimed at within-entity granularity.