
Unitree
Unitree H2 Plus
- Autonomy
- IIAssisted
- Pre-release
- 60/100Plausible
- Readiness
- Not assessed
- Price
- $100,000list

Unitree

RoboParty
Up to three
All humanoid robots. Every figure below is placed against the whole class, so a gap can be read against how much the class itself varies.
The conditions each robot was assessed against, side by side. Rows marked in gold are where these machines genuinely differ.
| Condition | Unitree H2 Plus | RoboParty RP1 |
|---|---|---|
| Where it operates | ||
| Home / residentialthey differ | PoorNo consumer software or home-safety certification is documented; CAN bus and RS485 interfaces target developers, not households. | DependsNo consumer home deployment has been announced; RP1 targets labs, classrooms and developer workspaces, not households. |
| Office / retail | DependsSold to labs and integrators, not as a turnkey commercial deployment; buyers must build their own task software. | DependsNo commercial deployment exists yet; RoboParty has described research and development use, not business operations. |
| Warehouse / factorythey differ | DependsShips as bare hardware; industrial use needs a buyer-built perception and task stack on the Jetson Thor module. | PoorRP1 is positioned as a research and development platform; no industrial deployment or ruggedization has been documented. |
| Primary tasks | ||
| Household chores | PoorNo manufacturer or third-party source documents H2 Plus completing a household chore on its own. | PoorNo household or chore-task capability has been documented; the public demo covered balance recovery only. |
| Social / care | PoorBuilt as an AI and robotics development platform; no conversational or social-interaction feature is documented. | PoorRoboParty has not documented any social-interaction or companion behavior for RP1. |
| Specialized job | GoodDual SharpaWave dexterous hands and a claimed 75 degrees of freedom target manipulation research, its documented focus. | GoodBuilt for robotics research and embodied-AI development, with an open stack intended for a team to specialize toward its own project. |
| Supervision needs | ||
| Fully autonomous | PoorNo autonomous task completion is documented; operating H2 Plus requires active development, not unattended use. | PoorNo autonomous, unsupervised operation has been documented; the only public demo ran on a staffed show floor. |
| With oversight | GoodMatches its intended role: a supervised research platform running models built on the onboard Jetson Thor compute. | GoodMatches RoboParty's own demonstrated use: a staffed IROS 2026 session with visitors directly interacting with the robot. |
| Teleoperated | GoodGMSL, CAN bus and RS485 interfaces support teleoperated or piloted operation, typical of humanoid R&D programs. | GoodFits a lab or classroom pilot project, which is the use case RoboParty is building the open stack for. |
| Timeline | ||
| Ready today | PoorNot sold as a finished deployable product; Unitree lists global shipping as planned for late 2026. | PoorNo units have shipped to anyone; nothing is available to use today. |
| Near-term (1–2 yrs) | DependsGlobal shipping is planned for late 2026; the hardware exists now, but a buyer must still build the software. | DependsRoboParty has named a Q4 2026 mass-production target, but with no firm date, price, or confirmed distribution to outside teams. |
| Research phase | GoodPositioned explicitly as a research humanoid and NVIDIA's Isaac GR00T reference platform, its documented use today. | GoodMatches the platform's stated purpose: an open research and embodied-AI development kit, demonstrated live at IROS 2026. |
A robot being poor for a condition is a fact about fit, not a fault. “Depends” means the record names a limit worth reading, not that the answer is unknown.
None of these has shipped, so they are assessed on track record, engineering plausibility and how the demonstrations match the claims.
Excluded as unevidenced, so it is left out of the number: RoboParty RP1
Never averaged into it. These describe how much the maker has disclosed, not what the machine does.
| Unitree H2 Plus | RoboParty RP1 | |
|---|---|---|
| ReadinessShip-date specificity, pricing transparency, distribution and regulatory status. Published beside the number; never averaged in. | 62/100 | 42/100 |
| DisclosureHow much of the product is disclosed. Higher means fewer fundamental unknowns remain. Published beside the number; never averaged in. | 35/100 | 32/100 |
A number only means something next to the spread it sits in. Each track is the real range across every classified robot in the class, with the median marked.
These robots share no measured specification with enough published figures to plot. That is a gap in the record, not a finding about the robots.
Written on each record, and the only thing here every robot on the site carries.
What each design gives up to be good at what it is good at.
NVIDIA’s own reference hardware is not the same claim as a robot shown completing a task on its own.
No source documents H2 Plus finishing a job without a developer building the control software first, even with local AI processing available on the onboard Jetson Thor module.
GMSL, CAN bus and RS485 tell you who this ships for: an integration team, not a buyer who wants to switch it on.
RP1 is open by design, but openness is also the only thing proven so far: the IROS demo showed balance recovery, not the manipulation or task-autonomy claims the fuller platform makes.
Building on RP1 today means building on a roadmap, because the motion-control system, simulation environment and SDKs RoboParty calls “full-stack” are still being opened in stages, not shipped complete.
Stated by the record itself. A robot being wrong for a household is a fact about fit, not a fault.
The upkeep ownership hands back to you, whatever the autonomy level says.
Classifications rest on different depths of evidence. A pre-release assessment and a verified one are not the same kind of claim.
| Robot | Evidence status | Sources reviewed | Last assessed |
|---|---|---|---|
| Unitree H2 Plus | Not Consumer-Ready | 1 | Sep 2026 |
| RoboParty RP1 | Pre-Release | 3 | Oct 2026 |
Evidence status counts the sources behind a classification, not their kind. A high count can be maker documentation and price coverage, so read it as how much was reviewed rather than as independent testing. How we weigh evidence.
The full recordUnitree H2 PlusSpecifications, evidence, the Fit Check in full, and every tracked change.
The full recordRoboParty RP1Specifications, evidence, the Fit Check in full, and every tracked change.Robovations classifies robots; it does not rank them. Nothing on this page is a recommendation, and no manufacturer influences a classification or its placement here.
Comparing across classes is allowed — the page will switch to per-class positioning and tell you it has.