Shipping ProHand: The Human-Like Robotic Hand Built for Dexterous Manipulation

Most robots still cannot use their hands.

That sounds strange, because robots have gotten dramatically better over the past decade. They can navigate warehouses, run, jump, perceive their surroundings, and reason over visual scenes. But when it comes to actually touching the world, most robots still rely on simple grippers.

That works if the object is rigid, predictable, and placed exactly where the robot expects it to be. It breaks down quickly when the task requires the kind of manipulation humans do every day: adjusting grip, feeling contact, using multiple fingers, handling soft objects, or recovering when something slips.

We started Proception because we kept running into the same bottleneck: humanoid robots will not become truly useful until they can manipulate the world with something much closer to a human hand.

Today we are introducing ProHand 1.0, a high-dexterity robotic hand built for researchers and robotics companies working on contact-rich manipulation.

https://youtu.be/S61xWpz7_qM

Why We Built a Human-Like Hand

The human hand is not just a gripper. It has more than twenty degrees of freedom, dense tactile feedback, tendon-driven actuation, and an incredible ability to adapt through contact.

That is what makes simple tasks deceptively hard for robots. Tying shoelaces, opening packaging, handling tools, plugging in cables, folding clothes, or repairing electronics are not just vision problems. They are contact problems.

To build ProHand, we worked closely with hand surgeons and designed the system around the structure of the human hand.

ProHand includes:

• Twenty-two degrees of freedom, with multiple joints per finger, enabling a wide range of human-like motions.
• Tendon-driven actuation, where motors pull cables to move the fingers while keeping the hand lightweight and compact.
• A compact and robust mechanical design, built to be impact tolerant, serviceable, and practical for real robotics research.
• Integrated skin-like sensors, which detect contact and support grip control during manipulation.

The goal was not to build a hand that looks human for aesthetics. The goal was to build hardware that can interact with the world in the same manipulation regime as human hands.

The Bigger Problem Is Data

Dexterous manipulation is not only a hardware problem. It is also a data problem.

AI has made enormous progress in vision and language because those fields had access to large-scale datasets. Robotics does not have the same advantage. Manipulation data is expensive, slow, and usually collected in labs through teleoperation.

Teleoperation is useful, and ProHand supports standard teleoperation workflows out of the box. But it has two big limitations.

First, it does not scale easily. Data collection is limited by how many robots you have and how much time operators can spend controlling them.

Second, it loses important human interaction signals. When someone teleoperates a robot, they are not directly touching the object. The robot sees motion commands, but it often misses the subtle pressure, contact, and adjustment strategies humans naturally use when manipulating objects.

Because ProHand is designed around human hand kinematics and skin-like sensing, we can also collect data directly from human hands.

That is why we built ProGlove.

ProGlove turns the same sensor skin used on ProHand into a wearable data collection system. A person can wear the glove, interact with real objects, and capture human manipulation data without needing a robot in the loop. Combined with a headset for visual context, researchers can collect real human hand interaction data and transfer those movements and skills to ProHand.

Our bet is simple: the fastest path to dexterous robots is to build hardware close enough to the human hand that we can learn from human hands directly.

Shipping This Week

Proception has raised $11 million in seed funding led by First Round Capital, with participation from Y Combinator and BoxGroup. This allows us to expand the team, scale production, and keep building the hardware and data infrastructure needed for dexterous manipulation.

Our team has designed consumer electronics at Apple that reached millions of users, built robotic systems at Tesla and early-stage startups, and spent years working on the messy details of real hardware: actuation, sensing, reliability, calibration, control, and data collection.

We are building ProHand in Mountain View, and the first batch of units is shipping this week to researchers and robotics companies.

If you are building dexterous manipulation systems, humanoid robots, teleoperation workflows, or robot learning pipelines, you can order ProHand on our website.

If you want to help build the hands that make humanoid robots useful, we are hiring https://www.proception.ai/careers.