Unitree packs more lab into H2 Plus

Unitree is presenting H2 Plus as a research and development version of its H2 humanoid, with a specification sheet now built around onboard compute and manipulation. The verified fact is on the company’s official page: the robot is listed with an NVIDIA Jetson T5000 module, 2,070 FP4 TFLOPS, 128 GB of unified memory, about three hours of battery life, five-finger tactile hands and 75 total body-and-hand degrees of freedom. This is not just another walking demo. It is a signal that Unitree wants the humanoid to become a more integrated development platform.

In a humanoid robot, degrees of freedom are the controlled axes of motion, such as a shoulder, wrist or finger joint. More axes can support richer movement, but they also make control harder. Unitree lists 31 joint motors for the body, 22 active degrees of freedom per tactile hand, maximum arm torque of 120 N.m and maximum leg torque of 360 N.m. It also gives a rated arm payload of 7 kilograms, with a peak figure of about 15 kilograms. Those numbers do not prove reliable dexterity in a factory or a home, but they define the physical envelope that control software will have to use.

The more important detail is the packaging of mechanics, perception and software workflow. The H2 Plus page ties the robot to NVIDIA Isaac GR00T, Isaac TeleOp, Isaac Sim, Isaac Lab, Isaac ROS and Jetson Thor. In practical terms, Unitree is not only selling a humanoid body. It is describing a hardware base for collecting demonstrations, training robot policies in simulation, testing them, and then moving them onto the robot. A robot policy is the control software that selects the next action from sensor input and a near-term goal.

That framing says something about where humanoid robotics is heading. The field is slowly moving from individual videos of balance, walking or dance toward the infrastructure needed to learn useful tasks: wide-view vision, wrist cameras, microphones and speakers, remote emergency stop, local inference hardware and enough interfaces for secondary development. Many open questions remain, including price, availability, durability outside controlled demos and safety around people. Still, the direction is clear. Manufacturers want humanoids to become repeatable experimental machines, where teams can compare models, hands, sensors and scenarios without rebuilding the whole platform for every trial.