This isn't a single robot you'll buy off a shelf. It's a detailed reference architecture that combines three key pieces into one unified development platform:

• The Body: Unitree H2 Plus — A human-scale chassis standing roughly 6 feet (1.8 meters) tall and weighing about 150 pounds (68 kg). It boasts 31 degrees of freedom (DoF) in the body, with powerful arm torque (up to 120 Nm) and leg torque (up to 360 Nm), supporting a 7 kg rated arm payload (15 kg peak) .
• The Hands: Dual Sharpa Wave — This is where the H2+ truly differentiates itself. The Singapore-based firm contributed its tactile, five-fingered hands, which add 22 DoF for dexterous manipulation. In total, the full robot platform has 75 degrees of freedom . These hands are designed for delicate, high-precision tasks most humanoid robots can't yet perform, from assembling electronics to administering an IV .
• The Brain: NVIDIA Jetson AGX Thor T5000 — The onboard computer is a powerhouse. It runs on an NVIDIA Blackwell GPU, delivering 2,070 FP4 teraflops of AI performance, paired with a 14-core Arm CPU and 128 GB of unified memory. All of this is pre-loaded with the NVIDIA Isaac GR00T open software stack, which includes foundational models, simulation tools, and workflows for data capture, training, and deployment .

The result is a “body” and “brain” that arrive together, solving a massive integration headache from day one .

Why an "Open Reference Design" Matters

The real story isn't just the hardware specs—it's the strategy to un-fragment a messy field. Today, a humanoid robotics lab might spend years sourcing a body, integrating hands, building a compute stack, and custom-coding a software pipeline before they can even start the real research: teaching a robot to do something useful .

The H2+ reference design aims to collapse that timeline. By offering an open, non-proprietary blueprint, NVIDIA and its partners give any research institution a shared, frontier-level starting point. The software is Isaac GR00T. The compute is Jetson Thor. The body and hands are already integrated. Researchers can immediately plug into a full-stack system for capture, simulation, training, and deployment .

Jensen Huang framed it as a platform for a new era of computing: "The NVIDIA Isaac GR00T Reference Humanoid Robot gives researchers a single, open platform to make breakthrough discoveries toward general-purpose physical intelligence" .

The first wave of research partners already reflects this ambition. Early adopters include Ai2, ETH Zurich, Stanford Robotics Center, and UC San Diego's Advanced Robotics and Controls Laboratory, with the first units expected to ship to researchers later in 2026 .

Standardizing the Road to General-Purpose Robots

The collaboration is also a significant geopolitical and industrial flex. It links a U.S. AI giant with a Chinese hardware leader and a Singaporean specialist, creating a single reference design that other companies can adopt, customize, and build upon . The hope is that this open blueprint will become a standard, helping the global humanoid industry scale faster by converging on a common hardware-software foundation.

In a world where proprietary humanoid platforms risk creating new walled gardens, the H2+ is a deliberately open bet that the fastest path to robots that can perform real work is to build a shared starting line, not a series of locked doors.

This isn't a single robot you'll buy off a shelf. It's a detailed reference architecture that combines three key pieces into one unified development platform:

• The Body: Unitree H2 Plus — A human-scale chassis standing roughly 6 feet (1.8 meters) tall and weighing about 150 pounds (68 kg). It boasts 31 degrees of freedom (DoF) in the body, with powerful arm torque (up to 120 Nm) and leg torque (up to 360 Nm), supporting a 7 kg rated arm payload (15 kg peak) .
• The Hands: Dual Sharpa Wave — This is where the H2+ truly differentiates itself. The Singapore-based firm contributed its tactile, five-fingered hands, which add 22 DoF for dexterous manipulation. In total, the full robot platform has 75 degrees of freedom . These hands are designed for delicate, high-precision tasks most humanoid robots can't yet perform, from assembling electronics to administering an IV .
• The Brain: NVIDIA Jetson AGX Thor T5000 — The onboard computer is a powerhouse. It runs on an NVIDIA Blackwell GPU, delivering 2,070 FP4 teraflops of AI performance, paired with a 14-core Arm CPU and 128 GB of unified memory. All of this is pre-loaded with the NVIDIA Isaac GR00T open software stack, which includes foundational models, simulation tools, and workflows for data capture, training, and deployment .

The result is a “body” and “brain” that arrive together, solving a massive integration headache from day one .

Why an "Open Reference Design" Matters

The real story isn't just the hardware specs—it's the strategy to un-fragment a messy field. Today, a humanoid robotics lab might spend years sourcing a body, integrating hands, building a compute stack, and custom-coding a software pipeline before they can even start the real research: teaching a robot to do something useful .

The H2+ reference design aims to collapse that timeline. By offering an open, non-proprietary blueprint, NVIDIA and its partners give any research institution a shared, frontier-level starting point. The software is Isaac GR00T. The compute is Jetson Thor. The body and hands are already integrated. Researchers can immediately plug into a full-stack system for capture, simulation, training, and deployment .

Jensen Huang framed it as a platform for a new era of computing: "The NVIDIA Isaac GR00T Reference Humanoid Robot gives researchers a single, open platform to make breakthrough discoveries toward general-purpose physical intelligence" .

The first wave of research partners already reflects this ambition. Early adopters include Ai2, ETH Zurich, Stanford Robotics Center, and UC San Diego's Advanced Robotics and Controls Laboratory, with the first units expected to ship to researchers later in 2026 .

Standardizing the Road to General-Purpose Robots

The collaboration is also a significant geopolitical and industrial flex. It links a U.S. AI giant with a Chinese hardware leader and a Singaporean specialist, creating a single reference design that other companies can adopt, customize, and build upon . The hope is that this open blueprint will become a standard, helping the global humanoid industry scale faster by converging on a common hardware-software foundation.

In a world where proprietary humanoid platforms risk creating new walled gardens, the H2+ is a deliberately open bet that the fastest path to robots that can perform real work is to build a shared starting line, not a series of locked doors.