Full-size humanoid transition brief · Canada · 2026
What changed for Canadian humanoid robotics teams?
Short answer: Unitree H2 is a new integration decision, not a drop-in H1-2 replacement. Both remain in the approximate 70 kg class and publish about 7 kg rated arm load, but H2 moves from 27 to 31 total joints, publishes a 3-axis waist and 2-axis head, increases battery energy from 0.864 to 0.972 kWh, and changes the published perception stack from H1-2's 3D LiDAR plus depth camera to a wide-field binocular camera. A custom-development team should specify H2 EDU and revalidate the delivered sensing, compute, hands, software and facility controls before migrating.
01 · The decision
The product changed. Your evidence must change with it.
H2 publishes a different joint layout and human-facing design, but the important buyer question is not “Is it newer?” It is “Which parts of our validated H1-2 workflow no longer transfer without proof?”
Product-image integrity: the two robot images above come directly from SpeedyDrone and Unitree. They have not been AI-redrawn, reconfigured or supplemented with invented hands, sensors or controls.
02 · One-view comparison
Compare the integration-relevant facts, not the silhouettes
The figures below come from Unitree's current official H2 and H1-2 pages. “Maximum,” “peak,” “rated” and “listed battery life” are deliberately kept separate because they do not describe the same test condition.
Source terminology is preserved. Published manufacturer values are not safe operating limits or a substitute for task testing. Unitree also notes that specifications may vary with scenario and configuration.
03 · Articulation
The four-joint increase is real; the one-to-one anatomy story is not published
H2's table gives a complete 31-joint allocation. H1-2's public table gives 27 total and explicitly identifies 12 leg joints plus 14 arm joints, but it does not label the remaining joint in the same breakdown. That means teams can defend the net increase, not an invented joint-by-joint substitution map.
12 leg + 14 arm + 3 waist + 2 head joint degrees of freedom.
12 leg + 14 arm are explicit. The official comparison table used here does not assign a label to the one remaining joint.
Engineering implication: H2's published 3-axis waist and 2-axis head layout can create different reachable poses and human-interaction possibilities, but it also changes joint indexing, collision geometry, motion limits, whole-body control and test coverage. Rebuild the joint map from the exact H2 model and software release instead of extending an H1-2 array by four entries.
04 · Integration deltas
Three changes that deserve separate validation plans
A procurement comparison becomes useful only when each specification turns into a testable engineering question.
1. Perception: a changed stack, not a universal winner
Unitree publishes a wide-field binocular camera for H2 and 3D LiDAR plus a depth camera for H1-2. These are different sensor arrangements; neither label proves better localization, obstacle coverage or manipulation performance in your scene.
Before migration, record required field of view, minimum and maximum working distance, lighting, reflectivity, depth error tolerance, calibration method, frame names, timestamps, data formats and compute load. Then run the same scene-level acceptance cases on the exact H2 configuration.
2. Battery: more energy is not a promised longer shift
H2's published battery is 0.972 kWh versus 0.864 kWh for H1-2, and Unitree lists about three hours for H2. The H1-2 comparison table used here does not give an hours figure, so a direct runtime uplift cannot be calculated from official data.
Motion profile, payload, compute, ambient temperature, battery condition, firmware and idle time all affect a real session. Define a duty cycle and log useful work, charging, swapping, cooldown and recovery instead of budgeting from capacity alone.
3. Manipulation: the rated arm-load class did not increase
Both pages publish about 7 kg rated arm load. H2's peak figure is about 15 kg, while H1-2's is about 21 kg. Peak is not a continuous payload and can depend on posture, configuration and test conditions; the lower H2 peak number does not prove lower performance for every task.
Acceptance must use the real hand, wrist, object, centre of mass, reach, orientation, speed and safety margin. Specify whether quoted payload includes the hand, adapter, tooling and cable load.
4. Compute and hands: order by configuration, not family name
H2 EDU adds the user-development computer and lists multiple high-compute and hand options. Those options are not evidence that every H2 EDU shipment includes the same module or dexterous hand.
Write the processor, memory, storage, accelerator, operating system, network access, mounting, power, hand model, handedness, tactile capability, cabling and interface entitlement into the bill of materials.
05 · Software reality
A familiar Unitree ecosystem does not make the port automatic
Unitree's current public SDK, model and XR teleoperation resources reference H2 and H1-2. That establishes an ecosystem path; it does not establish binary compatibility between two robots, two delivered configurations or two repository commits.
Geometry and joints
Confirm the exact H2 URDF or model package, link names, joint order, limits, collision meshes, inertias, end effectors and coordinate frames.
Messages and timing
Inventory DDS topics or other interfaces, message fields, rates, timestamps, watchdogs, network topology and loss behaviour.
Control and evidence
Re-run teleoperation mapping, balance transitions, perception, hand services, application state machines and failure recovery against frozen versions.
Record the robot serial number, delivered firmware, SDK version, repository URL and commit, model asset version, hand service, simulator version and operator device in the acceptance record. A screenshot of a repository name is not configuration control.
06 · Migration gate
Move from H1-2 evidence to H2 evidence in six controlled gates
The goal is not to reproduce every H1-2 demo. It is to prove the few capabilities your lab or solution actually depends on.
-
Freeze the H1-2 baseline
Archive hardware identifiers, hands, compute, sensors, calibration, firmware, SDK, model files, repository commits, network settings and representative test logs.
Write down which behaviours are validated and which are only demonstrations or operator impressions.
-
Define the H2 EDU bill of materials
Name the robot variant, user-development compute, accelerator, hands or grippers, batteries, charger, controller, cables, documentation and access rights.
Do not use “H2 EDU” as a substitute for an itemized delivered configuration.
-
Build the interface delta
Compare joint names and limits, coordinate frames, sensing, message definitions, timing, network requirements, compute environment, hand services and stop or recovery functions.
Assign an owner and evidence requirement to each unresolved difference.
-
Port into an isolated environment
Start with the exact H2 model and supported interface versions. Validate parsing, transforms, limits and state transitions before issuing motion commands.
Simulation can expose mapping errors, but it does not replace hardware and site acceptance.
-
Commission from low energy upward
Use controlled access, trained roles, conservative limits and pre-agreed stop and recovery procedures. Check communications loss, sensor loss, command timeout and restart behaviour.
Increase motion scope only after the preceding evidence passes.
-
Accept the real task
Run the agreed scene, payload, route or interaction with measurable success, repeatability, timing, intervention and failure-response criteria.
Record which hardware or software changes trigger partial or full re-acceptance.
07 · Procurement gate
For custom development, the model suffix matters
SpeedyDrone currently lists both H2 Basic and H2 EDU. Unitree's table lists secondary development for H2 EDU, not the standard H2. Keep this distinction short and decisive.
Evaluate as a platform configuration
Do not assume secondary-development access, user-development compute or a particular hand simply because the chassis is H2. Review the live SpeedyDrone H2 Basic page and obtain written confirmation for the exact intended use.
Start here for a custom software program
Unitree lists secondary development for H2 EDU and adds a user-development computer. That makes EDU the defensible quotation path for a team planning custom control, perception, simulation or teleoperation—subject to an exact bill of materials and interface confirmation.
Current SpeedyDrone pathways: review the complete Unitree H2 Series, check the live H2 EDU listing, and inspect the separate dexterous hands and grippers collection when a manipulation task requires a named end effector.
08 · Canadian acceptance
Turn the quotation into a configuration-control document
For a Canadian university, innovation lab or integrator, the purchase decision should define what arrives, what can be developed, what will be tested and who accepts the residual risk.
Exact hardware
- H2 variant and serialised bill of materials
- Compute modules, memory, storage and accelerator
- Sensor, hand, wrist, adapter and cable configuration
- Battery, charger, controller, spares and documentation
Development baseline
- Interface and secondary-development entitlement
- Firmware, SDK, model and repository versions
- Supported operating system and simulator path
- Examples, hand service, update and support boundaries
Acceptance evidence
- Perception and calibration cases in the real scene
- Task payload, posture, speed and repeatability
- Network loss, command timeout, stop and recovery
- Change-control and re-acceptance triggers
Commercial confirmation
- Current quotation and delivery scope
- Warranty terms and exclusions in writing
- Commissioning and integration responsibilities
- Escalation, service and documentation handover
Safety boundary: Unitree warns that humanoid robots have a complex structure and powerful output and instructs users to maintain sufficient safety distance. CCOHS guidance on robots and cobots calls for task-based risk assessment across integration, programming, operation and maintenance, along with suitable controls and testing. An EDU label does not certify a collaborative application; the facility and task still need their own review.
09 · Buyer-fit verdict
Choose the evidence path that matches your program
H2's strongest case is not that every number increased. It is that its current development platform, explicitly published 3-axis waist and 2-axis head layout, and changed HRI-oriented hardware serve a project you can define and revalidate.
Move toward H2 EDU when…
- the 3-axis waist, 2-axis head or current H2 development path matters to the research question;
- the team can validate a binocular perception stack in its real environment;
- the quotation names compute, hands, interfaces and software versions;
- the budget includes integration, commissioning and re-acceptance—not only hardware.
Do not call it an automatic upgrade when…
- the project depends on a validated H1-2 LiDAR and depth-camera pipeline;
- existing control or teleoperation code has not been mapped to the exact H2 interfaces;
- a specific hand, compute module or peak-load figure is assumed rather than quoted;
- the team has no measurable facility and task acceptance plan.
10 · Buyer questions
Frequently asked questions
Is Unitree H2 a direct replacement for H1-2?
No. H2 remains in the approximate 70 kg class but changes total joints, sensing, battery voltage and energy, compute packaging and development configuration. Treat it as a new integration baseline and revalidate the hardware, software, task and facility controls that matter to your project.
What changed from 27 DoF on H1-2 to 31 DoF on H2?
H2 publishes 31 total joint degrees of freedom: 6 per leg, 7 per arm, 3 at the waist and 2 at the head. H1-2 publishes 27 total and explicitly lists 6 per leg and 7 per arm, but the public comparison table used here does not label the remaining joint. The defensible statement is a net increase of four, not an invented one-to-one joint substitution.
Does H2 retain H1-2's 3D LiDAR and depth-camera sensing?
Not according to the current official parameter tables. Unitree lists a wide-field humanoid binocular camera for H2 and 3D LiDAR plus a depth camera for H1-2. Confirm the delivered sensor configuration and revalidate field of view, range, lighting, calibration, data interfaces and compute requirements for the real task.
Is H2 stronger than H1-2 for arm payload?
The official pages do not support that broad claim. Both publish about 7 kg rated arm load. H2 lists about 15 kg peak and H1-2 about 21 kg peak, but peak values are conditional and are not continuous working payloads. Validate the actual hand, object, posture, reach, motion and safety margin.
How do the H2 and H1-2 batteries differ?
Both publish 15 Ah. H2 lists 0.972 kWh and a maximum 75.6 V, while H1-2 lists 0.864 kWh and a maximum 67.2 V. Unitree lists about three hours for H2 but gives no H1-2 hour figure in the comparison table used here, so the capacity difference should not be presented as a guaranteed runtime uplift.
Which H2 version supports secondary development?
Unitree's current table lists secondary development for H2 EDU and does not list it for standard H2. A Canadian team planning custom control, perception, simulation or teleoperation should request an exact H2 EDU configuration and confirm interfaces, compute, software access and support in writing.
Can an H1-2 software or teleoperation stack move directly to H2?
Do not assume it can. Unitree's public ecosystem references both robots, but H2 changes joint allocation and sensing, and delivered firmware, SDK, models, message definitions, hands and compute can differ. Compare exact versions, rebuild the interface map and repeat simulation, bench and task acceptance tests.
What must a Canadian lab put in the H2 quotation and acceptance plan?
Specify the H2 variant, serialised bill of materials, sensors, compute, hands, batteries, charger, accessories, development entitlement, firmware, SDK and model versions, support scope and commercial terms. Define measurable perception, manipulation, communications-loss, stop, recovery and task tests, plus workplace risk review and change-control owners.
Verification record
Official sources checked August 2, 2026
- SpeedyDrone Unitree H2 Series — current Canadian H2 family route and Basic/EDU context.
- SpeedyDrone Unitree H2 Basic — live Canadian product reference and authentic H2 image source.
- SpeedyDrone Unitree H2 EDU — current development-oriented product route and primary quotation CTA.
- Unitree H2 official page — H2 dimensions, weight, joints, sensing, battery, load, compute, development and configuration tables.
- Unitree H1 and H1-2 official page — H1-2 dimensions, weight, joints, sensing, battery, load, compute and official manufacturer imagery.
- Unitree official open-source index — public routes to SDK and robot-development resources.
- Unitree XR teleoperation repository — current official public H2 and H1-2 ecosystem reference; not proof of drop-in project compatibility.
- CCOHS Robots and Cobots — Canadian task-based risk assessment, controls, testing and workplace integration guidance.
Configure the H2 around the task you must prove
Send SpeedyDrone your required joints and motion envelope, sensing scene, compute and hand requirements, software baseline, facility controls and measurable acceptance cases. Ask the quotation to name the exact H2 EDU hardware, development entitlement and support boundary. Confirm current commercial terms before ordering.
View H2 EDU and Plan a Configuration Review
