Technology Snapshot
A concise view of platform maturity and deployment footprint.
Sharpa is taking a manipulation-first route into general-purpose robotics. Rather than leading with a generic humanoid pitch, it built a high-performance hand and tactile-learning stack first, then used that foundation to move up toward a full-body robot and an integrated autonomy story.
Platform maturity, autonomy stack, and flagship-system specifications in one view.
A concise view of platform maturity and deployment footprint.
At NVIDIA GTC 2026, Sharpa said Wave had demonstrated in-hand manipulation skills trained in Isaac Lab and transferred into the real world, while North and CraftNet were also shown as part of the full stack. Sharpa also said it had joined NVIDIA Inception and that Wave had been adopted by NVIDIA GEAR for research in data-driven robot learning.
Sharpa is a Singapore robotics company building high-performance manipulation systems and humanoid platforms around tactile-first learning. Founded in 2024, the company started with the Sharpa Wave dexterous hand and then expanded into North, a full-body robot, while developing CraftNet, its vision-tactile-language-action model stack for fine manipulation.
Sharpa says it was founded in 2024 as an AI robotics company focused on ultra-high-performance robots and core components for future general-purpose robotic applications. The companys structure already spanned Singapore headquarters, a Mountain View business center, and Shanghai manufacturing R&D.
According to Sharpa, Wave was introduced in May 2025 as a 1:1 human-scale dexterous hand with 22 active degrees of freedom and a proprietary tactile array. The company says the hand entered mass production and began shipping in October 2025, giving Sharpa a commercial hardware foothold before the broader humanoid push.
On January 19, 2026, Sharpa said its newly debuted North humanoid played autonomous ping-pong rallies, assembled windmills, dealt cards, and demonstrated robot photography at CES 2026. The company used the same moment to introduce CraftNet, its vision-tactile-language-action model for fine manipulation, showing how Wave, North, and the learning stack fit together.
At NVIDIA GTC 2026, Sharpa said Wave had demonstrated in-hand manipulation skills trained in Isaac Lab and transferred into the real world, while North and CraftNet were also shown as part of the full stack. Sharpa also said it had joined NVIDIA Inception and that Wave had been adopted by NVIDIA GEAR for research in data-driven robot learning.