%page_title%

This is perhaps the greatest challenge facing modern robotics. We have taught robots to walk and dance, and soon we may teach them to think, but humanoid droids still lack hands that are as functional as human hands. Why is this the case, and what do we need to do to finally develop this technology?

Three years ago, in one of the stores of the popular Żabka chain in Warsaw, you could use a robot that made hot dogs. In most cases, the device worked efficiently, but there were some mishaps. In one of the recordings (shown below), we see how the robotic arm, ignoring the bun that had fallen off the platform, drops the sausage in the air and then serves it to the customer without bread. In yet another video, the problem was the opposite: the device failed to push the sausage into the bun and only served bread with sauce.

These amusing cases of robotic incompetence from Poland highlight a broader problem facing modern robotics. And that problem is hands. The hot dog robot used distinctive tongs to hold the bun or sausage. Wouldn’t it have done better if it had a five-fingered hand with touch sensors to precisely adjust the grip pressure? Of course it would. However, such a design element is currently only available to humans (and a few other animals). Creating a robotic hand is one of the greatest challenges in robotics today.

When will we have palms for robots?

When it comes to stable movement on two or four legs, robots are already doing quite well. One of the flagship examples is the impressive work of Boston Dynamics, whose two-legged Atlas robot was able to not only walk and run in test conditions, but also perform cartwheels, somersaults, crawl, and breakdance. Boston Dynamics focuses on developing the balance of robots, rather than precise elements such as hands. However, there is a consensus among companies and institutions that specialize in these specific parts of the “body” for androids: creating a good robotic hand is incredibly difficult.

We have known since 2021 that Tesla is working on a humanoid robot. Optimus has so far only been presented in the form of a few prototypes. In the fall of 2025, Elon Musk shared the fact that the biggest engineering challenge in building the devices is the hands. Despite these difficulties, the Tesla boss announces that the robots will be available for sale at the end of next year.

Will fully functional robotic hands really exist in less than two years? Industry experts are trying to dampen the enthusiasm. Kevin Lynch of Northwestern University (Evanston, Illinois) works at the university’s Center for Robotics and Biosystems. The institution is part of a federally funded consortium to develop robotic hands. In an interview with The Wall Street Journal, the scientist leaves no doubt:

We have set 10 years as the time frame in which we will achieve the dexterity, functionality, and usability that will allow us to do what humans do. It’s certainly not next year, says Kevin Lynch, Northwestern University.

Nathan Lepora, professor of robotics and artificial intelligence at the University of Bristol, shares a similar opinion. In an interview with the BBC, he bluntly calls Musk’s promises “nonsense” and also puts the horizon at 10 rather than 2 years.

The human hand, a true marvel of evolutionary engineering

There are as many as 30 muscles and 27 tendons in our hand. At the ends of our arms, each of us carries a true marvel of biological engineering, inside which more than 17,000 touch receptors constantly read signals from the environment. The brain then converts them (in cooperation with sight) into a perfect combination of bends, pressure, and the right grip so that we can, for example, hold a pen in our hands and write with it, pick strawberries without crushing them, or, as the author of this article did while writing it, type at a speed of about 40 words per minute.

Engineers are still unable to replicate this mechanism for use in robots. For now, certain solutions are being tested. The London-based company Shadow Robot has installed special motors in the forearms and metal tendons in its prototype. Such a device can be controlled remotely via special sensors connected to a human hand. The machine simply repeats the operator’s movements.

Shadow Robot has produced about 200 such robots, which are mainly used in laboratories. As the company’s director, Rich Walker, points out in an interview with the BBC, although they are used in practical environments (e.g., for mixing chemicals), the whole initiative is still more of a pilot and experimental nature:

It’s basically a development kit for dexterity research. You get this equipment, you see what you can do in terms of manipulation, and then you know what you want to build in a larger system, says Rich Walker, director of Shadow Robot.

Will artificial intelligence help engineers?

The rapid development of artificial intelligence in recent years has brought some improvements. Shadow Robot, in collaboration with Google DeepMind, has developed a three-fingered robotic hand called DEX-EE. Each finger has precision sensors that, with the support of AI, are able to read the environment around the device and, above all, control pressure. In the video below, DEX-EE manipulates a box of matches with great precision. Next to it, you can see a visualization of the changing pressure sensed by each sensor.

Apparently, DEX-EE has also mastered the ability to shake hands. This activity, which is trivial for humans, requires great precision from a robot, as well as the selection of the appropriate force and a dynamic response to the stimulus in the form of a human hand. The device manipulates eggs or inflated balloons without damaging them. For now, however, DEX-EE is more of a research tool than a solution ready for commercial implementation.

Advanced artificial intelligence algorithms can help robots use their hands with the right amount of force. The right balance of pressure, so as not to destroy the object being held on the one hand, and to hold it on the other, is trivial for humans, but machines have to learn it. And the methods of learning this can sometimes be unconventional. At the Fraunhofer IFF Institute in Magdeburg, a robot was tasked with tapping people on the shoulder to learn, based on their responses, how much force is painful to humans. As part of the study, the device had to tap volunteers approximately 19,000 times.

We are in the testing phase. Costs remain an issue

Costs remain an issue. The BBC writes about Kinisi, a Bristol-based company that develops robots for various industrial and research applications. Depending on the purpose, various types of devices, such as suction nozzles or powerful tongs, can be attached to the “arms” of their machines. Kinisi has also developed a prototype three-fingered hand that is supposed to resemble a human hand. According to company representatives, the result is satisfactory, but the production of such an arm attachment for their robot costs £4,000, while standard robotic pliers cost only £400.

Tests, pilots, experiments. High costs and technological problems. In this situation, it is really hard to believe that a real robotic hand can be produced in the near future. However, it is possible that… it does not have to be created at all. What if a much better and more convenient solution has been with us for years?

Vacuum grippers instead of robotic hands

Berkshire Grey is an American company that provides robotics and AI solutions for industry. The company’s chief scientist, Matthew T. Mason, argued on the company blog that, for industrial use, robots with human-like hands will never replace devices equipped with vacuum grippers.

The principle is simple: a nozzle sucks in air, thereby attracting the object being lifted to the surface of the gripper. Mason points out the simplicity of this solution: such a “hand” has only one “finger,” no tendons, is equipped with a single pressure sensor, and is usually powered by one or two motors.

In simpler devices, it is difficult to achieve good suction on the surface of the object being lifted if it is porous or uneven. However, there are specialized grippers with foam or bellows suction cups for which this is not a problem. Mason gives the example of Berkshire Grey grippers, which are able to lift pens or shower washcloths without any problem.

Vacuum grippers are already standard in the industry

Mason compares foam suction cups to wheels. Humankind has been using wheels to do the work of legs for centuries, but for some time now, efforts have been underway to create robotic legs.

Will legs replace wheels? In some very specific areas, perhaps. Several humanoids equipped with legs are already working in warehouses. But will legs widely replace wheels? No chance. Wheels are one of the most valued inventions in human history—and for good reason. They are simple, lightweight, durable, and inexpensive, just like vacuum grippers. They require a minimum number of motors and sensors—just like vacuum grippers. The arguments in favor of wheels are strikingly similar to those in favor of vacuum grippers. Wheels will be with us forever. So will vacuum grippers, says Matthew T. Mason, Berkshire Grey.

We could also see vacuum grippers in the hot dog-making robot that started our story. In addition to a simple gripper, two suction cups were mounted on the arm of the device. They were used to lift and place a small paper bag on the bun. The aforementioned Kinisi also offers arms ending in suction cups for its robots.

A hand for a robot is still a thing of the future

One might get the impression that, given the convenience of vacuum grippers, creating a copy of a human hand for a robot is pointless. Of course, this is not the case. Research into the creation of such a component could lead us, for example, to the development of better prostheses for people who have lost their hands. Such a solution would also be potentially better than suction cups in tasks requiring greater precision.

However, the creation of robotic hands shows us above all how unique the human body is. Almost all of us have a pair of highly precise, universal devices that engineers have been trying to replicate for decades. Will they finally achieve their goal? We may find out at the end of next year, when the first robots from Tesla go on sale.