How Do H2 and G1 Differ in Humanoid Research Design?

Author : Toborlife AI | Published On : 05 Oct 2026

How Does Physical Scale Change Humanoid Research?

Physical scale changes the mechanics of a humanoid experiment. H2 stands 1.82 m tall and weighs about 70 kg with its battery. G1 stands 1.32 m tall and weighs about 35 kg, while some G1 EDU configurations weigh slightly more.

That difference affects more than floor space. A taller and heavier robot has different limb inertia, mass distribution, reach, contact forces, and recovery dynamics. Those properties influence locomotion control, balance experiments, manipulation, and the equipment needed to support the robot during testing.

Researchers should therefore avoid treating a walking policy or motion controller developed on one platform as directly transferable to the other. A controller validated on G1 must still be evaluated against H2's body dimensions, joint limits, actuator behavior, and dynamic model before it can be considered suitable for H2.

Scale also changes the laboratory operating envelope. Unitree warns users to maintain sufficient distance from H2 because of the robot's mechanical complexity and power. A 70 kg humanoid falling or moving unexpectedly creates a different risk profile from a 35 kg platform, so laboratories should define operating boundaries, supervision procedures, and recovery methods around the exact robot being tested.

H2 becomes particularly relevant when the research question depends on adult-scale reach, body height, or interaction with spaces designed around human dimensions. G1 can be more practical when compactness, repeated experimentation, or a smaller test area matters more than full-height embodiment.

How Do H2 and G1 Differ in Joint Architecture?

The total degree-of-freedom count does not tell the full story. Researchers need to know where those joints are located and whether they contribute to locomotion, torso motion, wrist movement, or hand manipulation.

The unitree robotics h2 platform uses 31 body joints. Unitree specifies six degrees of freedom in each leg, seven in each arm, three at the waist, and two at the head. Both the base H2 and H2 EDU use this 31-joint body architecture.

The base G1 uses 23 joints. G1 EDU configurations can range from 23 to 43 depending on optional waist, wrist, and hand hardware. Each leg has six DOF and each arm has five before additional wrist or hand options are included.

This distinction matters because a larger numerical DOF count does not automatically mean a robot offers better whole-body motion. Optional hand joints may improve dexterity without changing leg kinematics. Additional waist joints may increase torso articulation without adding grasping capability.

Research teams should map each required motion to the joints responsible for producing it. A whole-body locomotion project may value waist articulation differently from a dexterous manipulation study, while an imitation-learning project may require both.

How Do Manipulation Requirements Change the Choice?

Manipulation research exposes one of the clearest hardware differences between H2 and G1.

Unitree rates each H2 arm for approximately 7 kg of rated payload and about 15 kg peak. The base G1 is listed at approximately 2 kg per arm, while G1 EDU is listed at about 3 kg. These figures provide an important reference when experiments involve carrying, reaching with objects, or applying forces through the arms.

Payload alone does not establish manipulation performance. A research task may also depend on wrist articulation, hand type, tactile sensing, object geometry, perception, motion planning, and force control.

The base H2 does not list a dexterous hand as standard, while H2 EDU supports multiple optional dexterous-hand models. Toborlife's H2 research configurations expand those choices further with several grippers and dexterous hands, including tactile and non-tactile options.

G1 follows a similar configuration-dependent pattern. The base G1 does not include the optional dexterous-hand DOF shown for G1 EDU. G1 EDU can add force-controlled three-finger hands, additional wrist joints, and other configuration-specific end effectors.

For grasping or physical-interaction research, the relevant comparison is therefore not simply H2 versus G1. It is the exact body, wrist, hand, sensing, and control configuration required by the experiment.

What Changes When Research Moves From G1 to H2?

Moving from G1 to H2 changes both the experiment and the infrastructure surrounding it.

H2's larger footprint and approximately 70 kg mass increase the space required for locomotion trials, recovery, transport, and safe separation. Research involving external equipment or human interaction also needs to account for H2's greater reach and arm payload.

Compute architecture differs as well. Unitree lists an Intel Core i5 for H2 platform functions and an Intel Core i7 for custom development on H2 EDU. H2 EDU can also support additional high-performance compute modules. G1 uses an 8-core high-performance CPU as its base compute platform, with higher-compute development options available in EDU configurations.

Development access is another critical dividing line. Unitree lists secondary development for H2 EDU but not base H2. The same pattern appears in the G1 line, where secondary development is listed for G1 EDU but not the standard G1.

That means a research team should not choose between the two platforms based on body dimensions alone. A robot may have the mechanical properties an experiment needs but still be the wrong configuration if the required control interfaces or development tools are unavailable.

Data and software should also be treated as platform-specific until validated. Research results from G1 can inform H2 work, but code, learned policies, calibration values, and motion constraints may require adaptation because the robots differ mechanically and computationally.

How Should Laboratories Choose Between H2 and G1?

Start with the requirements that software cannot change.

If the research depends on adult-scale reach, higher arm payload, a 1.82 m body, or interaction with human-sized workspaces, H2 provides physical characteristics that G1 does not. If the project prioritizes a smaller platform for repeated locomotion, manipulation, or learning experiments, G1 may fit the laboratory more efficiently.

Then evaluate the exact development configuration. Base H2 and base G1 should not be assumed to provide the same secondary-development access as their EDU counterparts. Hand options, compute modules, wrist articulation, and total DOF also vary by configuration.

Toborlife currently offers several H2 configurations, including Basic, Edu Smart, Plus Standard, and Plus Ultimate, with different hand and computing options. Its G1 catalog likewise includes multiple EDU and manipulation-oriented configurations rather than a single research specification.

The purchasing decision should therefore begin with four questions: what body scale does the experiment require, what joints must be controlled, what physical interaction must the robot perform, and what development access is necessary to implement the research.

The total investment also includes laboratory space, safety controls, computing infrastructure, development work, operator training, and maintenance. Those requirements can differ substantially between a 35 kg compact humanoid and a 70 kg full-height system.

For research teams evaluating human-scale locomotion, manipulation, and embodied AI, explore the H2 configurations available at Toborlife AI and compare the exact hardware, hand, compute, and development options against the requirements of the research program.