The Robot Hand That Grips Like a Human's

A Hand Designed Around Human Anatomy

Most industrial robots grip objects with two rigid fingers or a simple pincer. That design works for uniform, predictable objects like boxes on a pallet, but it struggles with anything delicate, irregularly shaped, or variable in weight, the kind of object a human hand adjusts to instinctively. Mimic Robotics, a physical AI company based in Zurich, Switzerland, built the mimic hand M1 to solve that specific problem, describing it as the first robotic hand purpose-built for industrial automation.

The hand's joints replicate the functional range of a human hand, including finger abduction and an opposable thumb, spread across 15 actuated degrees of freedom and 21 joints in total. Rather than placing motors inside the hand itself, Mimic drives the fingers through tendons connected to actuators housed in the forearm, mirroring how a human hand is actually built, with muscle mass concentrated in the forearm and tendons doing the work at the fingertips. The company states this keeps the hand lighter and more durable for sustained industrial use.

What the Hand Can Sense and Hold

The M1's specifications describe an unusually wide range for a single gripper. It maintains a steady hold on payloads exceeding 25 kilograms, about 55 pounds, while simultaneously detecting forces as small as 0.1 newtons, sensitive enough to register the difference between a firm grip and one about to crush something fragile. Tactile sensors built into the fingertips supply that force feedback directly, and because every joint functions as both an actuator and a sensor, the hand continuously reports detailed information about what it is touching, not only where its fingers are positioned.

In footage released by the company, the hand used a pair of tweezers to place a small component onto a circuit board and tap it into position, then separately picked up a bolt and passed it between two of its own fingers to a second hand. Both are modest actions, but each depends on precise force control and timing, the same qualities that simple two-finger grippers typically lack.

Training Data Without a Robot in the Room

This kind of hand still has to be taught, and Mimic addressed that through a companion device called the Mimic Wearable U1. It is a passive exoskeleton that a person wears while performing a task with their own hands. A rigid linkage constrains the wearer's hand to only the motions the M1 can physically perform, so the resulting data maps directly onto the robot rather than requiring translation afterward.

This matters because the standard method for collecting robot training data, teleoperation, requires a physical robot on site and struggles to scale. With the wearable, a person can generate usable training data simply by doing the task, without a robot present at all. Mimic designed the hand and the wearable together specifically so the human hand stays a fixed reference point across both the hardware and the data used to train it.

Where the Hand Is Meant to Work

The finished system reaches customers through configurable robot stations, available with one or two arms mounted on either a stationary table or a mobile platform. Mimic points to three main uses: complex manual assembly and kitting in manufacturing, packaging and sorting of irregularly shaped objects, and broader manipulation tasks in environments where object position, shape, and material vary constantly, conditions that conventional fixed automation typically cannot handle.

Why Robotic Hands Improved More Slowly Than Arms

Industrial robots have made real progress in reach, speed, and repeatable motion over the past several decades. Robotic hands improved more slowly. Two-finger grippers work for uniform, rigid objects, but they cannot easily adapt to the range of shapes, weights, and fragility found across a typical factory or warehouse. That gap meant many delicate manual tasks still had to be done by people, not because a robot could not reach the object, but because nothing at the end of its arm could handle it appropriately.

The M1 targets that specific problem rather than automation broadly. Whether it succeeds at scale depends on real deployments still ahead, and its usefulness will depend as much on the software trained to control it as on the hardware itself. What it represents so far is a hand engineered to match a wider range of human grip behavior than a fixed gripper allows, aimed at manufacturing and logistics tasks that have remained manual specifically because existing robotic hands could not handle them.