Humanoid robots are getting smarter brains, stronger bodies, and now, better hands.
Sharpa announced that its Wave tactile five-finger robot hands are being used in the NVIDIA Isaac GR00T Reference Humanoid Robot, making it the first dexterous humanoid reference design on the Isaac GR00T development platform to include Sharpa’s hands for tactile manipulation. The setup combines a Unitree H2 humanoid body, Sharpa Wave hands, NVIDIA Jetson Thor onboard computing, and Isaac GR00T software workflows into one integrated research platform.
That may sound like a parts list for a very expensive robot science project, but the point is simple. NVIDIA and its partners are trying to give researchers a humanoid system that can move, sense, train, and deploy skills faster without spending days just getting the hardware to behave. Sharpa says the validated configuration can reduce setup time from days to hours.
The big story here is touch. A humanoid robot that can walk across a room is impressive, but a humanoid that can actually handle objects without crushing, fumbling, or dropping them is much more useful. Real work usually happens through hands. Turning a knob, picking up a tool, sorting parts, carrying objects, or working with another robot all require dexterity and feedback.
Sharpa’s Wave hands bring 22 degrees of freedom per hand, giving the full robot 75 total degrees of freedom across the body and hands. Each fingertip includes a high-resolution digital tactile array with more than 1,000 pixels per fingertip and 0.02N pressure sensitivity. In human terms, the robot is not just closing a claw and hoping for the best. It is getting pressure feedback that can help it understand how it is touching something.
That is important for the next stage of humanoid robotics. The industry has already produced plenty of impressive walking, dancing, and backflipping videos. The harder challenge is fine manipulation. Robots need to learn how objects feel, shift, slip, resist, and respond in real time. A hand that can sense pressure gives AI models more useful data to work with.
NVIDIA’s reference design is also meant to smooth out the development pipeline. Researchers can capture upper-body demonstration data with Isaac Teleop, simulate and test manipulation policies in Isaac Sim and Isaac Lab, then deploy those policies on the robot using Jetson hardware for real-time inference and control.
Sharpa founder David Li said the company’s goal is to make robots “genuinely productive” by improving fine manipulation through tactile hardware and AI models. NVIDIA robotics product director Spencer Huang said dexterous hands are essential for humanoids to perform useful manipulation tasks in the real world.
For Clanks readers, the takeaway is that humanoid development is moving past the legs. Walking is still hard, but hands may be where the real money is. A robot that can stroll around the lab is fun. A robot that can pick up the right object, hold it gently, adjust its grip, and do the job without launching someone’s coffee mug across the room is a lot closer to being useful.
The humanoid race is no longer just about who has the coolest robot body. It is about who can give that body the sensors, hands, models, and workflows needed to actually work. In other words, the robots are finally learning that touching stuff is harder than it looks.
Join the Discussion