Singapore robotic hands firm Sharpa joins Nvidia and Unitree on humanoid robot project

TAIPEI – A future where robots can be deployed en masse to perform delicate tasks on manufacturing lines or in hospitals to care for the sick has come closer to reality, with a collaboration between Singapore-based robotic hands company Sharpa, computing giant Nvidia and Chinese robot maker Unitree.

They are collaborating to roll out humanoid robots in late 2026 to researchers, to help them train robot brains for tasks requiring precise hand movements – such as assembling a computer or administering an IV drip, a tall order for robots.

The tie-up was announced by Nvidia chief executive officer Jensen Huang in a highly anticipated keynote speech on June 1, on the sidelines of the Computex conference in Taipei, Asia’s biggest tech expo.

“Humanoid robots will bring physical AI to the world’s largest industries, opening a multitrillion-dollar economic opportunity,” said Huang, during a keynote at the Taipei Music Centre that drew a packed audience and was broadcast to 70 watch parties across Taipei.

The NVIDIA Isaac GR00T Reference Humanoid Robot or H2 Plus – which features Sharpa’s 22-degree-of-freedom robotic hands designed to mimic the dexterity and tactile sensitivity of human hands – is also the Singapore-headquartered company’s biggest collaboration to date.

Nvidia’s vice-president of physical AI simulation Rev Lebaredian said that Sharpa’s robotic hands allow researchers to train for tasks that require grasping, handling, adjusting and using objects with greater precision.

“That matters because the ‘brain’ of a humanoid robot is not just about understanding the world – it has to learn how different actions feel and behave on real hardware. The hands give researchers the data and physical feedback needed to connect robot learning with manipulation skills that can transfer to real-world tasks,” said Lebaredian.

Sharpa go-to-market global vice-president Alicia Veneziani said that with human-like hands, many robotics use cases can be unlocked.

“Precise assembly, food preparation, cleaning operations, even ironing a shirt – those are the kind of things that, without very precise hands and in particular, tactile sensing, (robots) won’t be able to do,” said Veneziani.

Sharpa was founded in 2024, and the company conducts research and development in Shanghai and has business operations in California in the US. The company previously collaborated with Nvidia to collect training data for the chip giant’s robotics foundational models.

Earlier in 2026, Sharpa’s robotic hands turned heads at the CES in Las Vegas when it demonstrated that it was able to deal blackjack cards and assemble pinwheels.

Those hands form just one part of the new humanoid robot. The robot also incorporates Unitree’s human-sized H2 humanoid robot body, which stands about 1.83m tall and weighs 68kg, as well as Nvidia’s Jetson Thor, the robot’s AI brain.

Other than the hardware, researchers will also gain access to Nvidia’s suite of AI models, simulation tools and software for training and operating humanoid robots.

“We built this for higher education and university researchers, because for them to build this is insanely hard to do,” said Huang.

Nvidia’s Lebaredian said that one of the biggest hurdles facing researchers is that today, they often have to piece together various robot hardware, software and computing systems themselves.

“These robots are kind of Frankenstein robots... So the labs spend a lot of time and energy making the basics of the robots work before they can start doing their research,” said Lebaredian, adding that the newest robot takes frontier humanoid research beyond the domain of the world’s largest tech companies and AI unicorns, and within reach of every lab.

Early adopters of the new humanoid robot include US AI research institute Ai2, Swiss engineering university ETH Zurich, Stanford University’s robotics research centre and the University of California San Diego’s Advanced Robotics and Controls Laboratory.

The H2 Plus will be available from Unitree, said the company, although further details have yet to be released.

During Huang’s two-hour-long keynote, he highlighted the industry’s shift towards agentic AI – technology capable of carrying out tasks and making decisions with minimal human intervention.

He argued that the growing use of AI agents will require new computing infrastructure and positioned Nvidia to provide the hardware and software underpinning it.

To that end, the US$5 trillion ($6.7 trillion) company unveiled its Vera central processing unit (CPU) chip for data centres and introduced a new generation of AI PCs powered by its new RTX Spark chip, which is designed to run AI models and agents locally on devices.

Computex runs from June 2 to June 5, with keynotes from semiconductor industry leaders including Intel chief executive Lip-Bu Tan, Qualcomm chief executive Cristiano Amon, Marvell chief executive Matt Murphy and NXP Semiconductors president Rafael Sotomayor.