This Robotic Hand Lifts Cans With One Finger and Feels a Feather

Most robotic hands can grip something or sense something, but rarely both well at once. A hand strong enough to clamp down hard usually loses the fine touch needed to feel something light. A hand sensitive enough to feel that touch usually can't move fast enough to be useful. Sharpa built a hand meant to do both.

The Sharpa Wave is a dexterous robotic hand designed to match a real human hand almost exactly — in size, proportion, and how it senses the world through touch. It carries 22 active degrees of freedom and more than 1,000 tactile sensing points concentrated at the fingertips, built by Singapore-based Sharpa for humanoid robots, research labs, and industrial integration.

Why Most Robotic Hands Fall Short

The human hand grips with real force, manipulates with precision, and senses texture without looking — three things rarely achieved together in robotics. Industrial grippers solved strength decades ago, clamping hard with no sense of touch and little resemblance to a human hand's shape, making them useless for delicate tasks like picking up an egg without crushing it.

Research-grade dexterous hands got closer on precision but historically struggled to pair that with real strength and speed. Sharpa describes combining over 20 newtons of fingertip force with movement above 4 Hz across all gestures as a trade-off long considered impossible to fully solve.

How Sharpa Wave Actually Works

The proportions aren't arbitrary. Wave is built at true 1:1 scale, following the same golden ratio — roughly 0.618 palm width to hand length — found in real hands. That means it can pick up and use the exact same tools and objects designed for human hands, without custom alternatives.

Underneath sit 22 active degrees of freedom, mirroring how human fingers and thumb articulate independently — a sharp step up from simpler grippers that compress multiple movements into a handful of motors.

The sensing layer is Sharpa's Dynamic Tactile Array (DTA) — over 1,000 tactile points concentrated at the fingertips, powered by a proprietary neural network algorithm. It's sensitive enough to detect feather-light contact, letting the hand adjust grip in real time rather than relying on pre-set force limits. Sharpa demonstrates this by lifting multiple soda cans on a single fingertip, and separately, opening and closing four times per second — strength and speed shown together, not traded off.

Built to Survive Real-World Use

Sharpa stress-tested Wave well beyond typical robotics demos: over 2,500,000 press cycles, more than 4,000 meters of friction travel, and an automatic protective clench that engages in 0.10 seconds to guard against sudden impact. The hand also passed thousands of mechanical shock cycles at 30g and 1,000+ hours of continuous operation across extreme temperature swings.

Who It's Built For

Wave ships with ROS2 packages and Isaac Sim simulation assets, designed to plug directly into existing robotics pipelines. That targets three groups: researchers training manipulation policies through reinforcement learning, teams collecting teleoperation data for imitation learning, and robot manufacturers integrating a precision end-effector into their own humanoid platforms.

Sharpa has already integrated Wave into NVIDIA's Isaac GR00T reference humanoid robot, signaling it's meant to plug into the broader humanoid ecosystem rather than stand alone.

Part of a Larger Platform

Wave is one piece of three. Sharpa North is the company's full humanoid robot, pairing whole-body control with Wave's manipulation skills for autonomous real-world operation. CraftNet is Sharpa's embodied AI model — a Vision-Tactile-Language-Action system built from neural networks running at different speeds, including a fast "System 0" reflex layer for instant reactions to unexpected contact. At CES 2026, Sharpa showed CraftNet running live autonomous fine-manipulation demos, pairing Wave's hardware with CraftNet's decision-making.