Carnegie Mellon researchers demo robotic hand costing less than $2,000

With prices ranging from $16,000 to $150,000, the cost of robotic hands has stunted progress in manipulation research.

Even as machine learning and artificial intelligence enable robots to acquire more skills and accomplish more tasks, the lack of suitable, reliable and affordable hardware has limited the transfer of these gains out of simulation and into the real world.

For example, robotic hands that can mimic a human’s more closely than the more affordable grippers currently used in labs can operate the same tools as people and accomplish more tasks.

Roboticists at Carnegie Mellon University’s School of Computer Science have solved this problem by developing a low-cost robotic hand that outperforms its more expensive counterparts.

The Low-Cost, Efficient and AnthroPomorphic (LEAP) Hand costs less than $2,000 and can be assembled in less than four hours using only a 3D printer and a screwdriver. (See video below.)

Deepak Pathak, an assistant professor at the Robotics Institute whose lab developed the hand, says: “Dexterous manipulation is a long-standing challenge in robotics.

“LEAP Hand significantly reduces the cost to perform research around this topic and could lead to major breakthroughs that allow robots to interact with objects in people’s homes, factories and warehouses.

“We have already demonstrated a variety of tasks with the open-sourced LEAP Hand that researchers previously accomplished with hands costing 10 to 100 times as much.”

LEAP Hand’s innovative design allows it to move in ways a human hand cannot. For instance, a human can flex its fingers in and out when its hand is open but not when closed in a near fist. This same motion is no challenge for LEAP Hand thanks to a motor located where the fingers join the palm.

This unique kinematic design gives LEAP Hand maximum dexterity regardless of finger poses. Its fingers and thumb independently move up and down, left and right, and rotate, allowing it to complete difficult tasks.

For robots to pick up silverware or dishes, fold laundry, open and close doors and drawers, flick switches, turn knobs, or fetch a drink from the refrigerator, they must successfully grasp and manipulate objects.

Humans use their hands – 27 bones, 27 joints, 34 muscles, more than 100 ligaments and tendons, countless blood vessels and nerves and, of course, the brain – to perform these tasks. LEAP Hand consists of far fewer parts.

The researchers have also released open-source software that enables the LEAP Hand to mimic the movements of a human hand during teleoperated trials, as well as to train the robot in simulation. It can also complete tasks learned from internet videos on sites like YouTube and apply those skills in the real world.

Kenneth Shaw, a PhD student in robotics, says: “We designed this hand so anyone interested in robotic manipulation research could build one and use it for their projects.

“LEAP Hand outperformed its competitors at a fraction of the cost.”

The researchers tested LEAP Hand’s ability to grasp objects like a soup can, mustard bottle, mug, baseball, drill and pan. They had it perform manipulation tasks such as picking up toys, opening drawers, stacking cups and rotating boxes.

Through all the testing, LEAP Hand proved to be stronger, more durable, and better at grasping and manipulating objects than similar but more expensive hands.

The researchers first demonstrated LEAP Hand at the Robotics: Science and Systems Conference this past July in Korea. Interest has been high since, with researchers from universities and tech companies large and small building LEAP Hands for their own robot hand applications and research.