Sharpa Showcases at GTC with NVIDIA, Advancing a New Paradigm for Robot Training Through Breakthroughs in Simulation

SAN JOSE, CA., March 18, 2026 — At NVIDIA GTC 2026, Sharpa showcased the stable real-world performance of its dexterous robotics technology and, on the opening day of the conference, was prominently mentioned by NVIDIA founder and CEO Jensen Huang during his keynote.

This showcase sends a clear signal: robotics is moving from being programmed to execute tasks toward learning from human data, and Sharpa is positioned at an important point in this technological shift.

In the live demonstration, Sharpa's dexterous robotic hand, Wave, demonstrated advanced tasks including in-hand manipulation. These fine manipulation skills have long been regarded as among the most challenging capabilities in robotics. More notably, these capabilities were developed through training on the NVIDIA Isaac Lab simulation platform and were stably transferred to a real-world system.

While many robotics approaches remain primarily vision-based, Sharpa is focused on tactile-driven dexterous manipulation. This capability is regarded as a key threshold for robots to perform complex real-world tasks.

Reliable sim-to-real transfer remains one of the central bottlenecks in scaling robotics for real-world deployment. Sharpa's progress suggests that this bottleneck is being systematically overcome.

Sharpa's technical path is built on two core capabilities. The first is large-scale skill training in simulation environments. The second is the training of its vision-tactile-language-action model (VTLA) using video and tactile data, enabling robots to learn human manipulation patterns and achieve a higher degree of autonomy.

To support this training paradigm, Sharpa has jointly developed the Tacmap simulated tactile system with NVIDIA as key infrastructure for tactile-driven robot learning.

At GTC, Xuezhou Zhu, Vice President of Research at Sharpa, further elaborated on this technical path: by introducing tactile signals into the learning loop through a high-degree-of-freedom dexterous hand, Sharpa is able to enhance complex manipulation capabilities, improve data efficiency, and significantly narrow the sim-to-real gap.

At the same time, Sharpa Wave has been adopted by NVIDIA GEAR for research in data-driven robot learning. Together, they have validated an important path: robots can learn complex manipulation capabilities directly from large-scale human video data and execute them stably in real-world systems.

This capability is built on Sharpa's anthropomorphic hand design. Wave features a 1:1 human hand size and configuration and is equipped with tactile sensing, allowing human operational data to be mapped more directly to robotic systems. This significantly reduces the embodiment gap between humans and robots.

This means robots no longer need to rely on expensive and difficult-to-scale robot data. Instead, they can be trained using more accessible and larger-scale real human data, thereby establishing a sustainable and scalable path for acquiring capabilities.

At GTC, Sharpa also showcased its full technology stack, including the North humanoid robot, the Wave dexterous robotic hand, and CraftNet, its model system that integrates vision, tactile sensing, and language. This integrated hardware-software capability enables Sharpa to form a closed loop from data and models to execution.

Sharpa also announced that it has joined the NVIDIA Inception Program and has received both the 2026 iF Product Design Award and the CES Innovation Award.

About Sharpa

Founded in 2024, Sharpa is an AI robotics company focused on building ultra-high-performance robots and core components for future general-purpose robotic applications. Sharpa's mission is to manufacture time by making robots useful.

Sharpa's global headquarters is in Singapore, with a business operation center in Mountain View, USA, and its manufacturing R&D center in Shanghai, China.