One hand is more expensive than an entire person. Why are manual crafts so difficult?

Is a pair of dexterous hands really that important?

On a 40°C summer day, curling up on the sofa in an air-conditioned room, opening a can of Coke, and watching a movie is simply the ultimate human enjoyment.

However, such a common action that everyone can do, opening a soda can, is an impossible task for robots with round hands.

As an inevitable path of development, how to equip humanoid robots with hands has always been the focus of research and development. If future robots don't have dexterous hands, it's already half out of character for human cosplay.

But the question arises: With so many people spending so much money over so many years, what's the actual situation of dexterous hands?

01

We consulted a lot of materials and interviewed experts in the academic circle. We found that although a dexterous hand seems like a small component, there are actually quite a lot of intricacies inside.

Although dexterous hands and humanoid robots seem to have become popular only in the past two years, in fact, related research has never stopped since the last century.

From the review paper "Dexterous hand towards intelligent manufacturing: A review of technologies, trends, and potential applications"

We found a paper from 1986 and discovered that the hand from forty years ago actually looks no different from the current ones in appearance.

People at that time were already imagining that robots equipped with hands could replace them as hard - working laborers.

From the 1986 paper "Design of the Utah/M.I.T Dextrous hand"

However, it's easy to make a prototype, but the technological iteration is much more complex. All subsequent evolutions share the common goal of making robotic hands more flexible, all - around, versatile, and more like human hands.

Dr. Wang from the Institute of Automation of the Chinese Academy of Sciences said that to make a good dexterous hand, one mainly has to put in a lot of effort in materials, sensors, and drives.

First of all, the improvement of materials can make dexterous hands both rigid and flexible.

It can be said that previous hands were all "tough guys made of steel." Asking them to do some delicate operations or pick up fragile items was simply a gamble.

To make the hands more flexible and versatile, the structural materials should pursue lighter weight and higher strength. For example, a titanium alloy skeleton wrapped in soft skin can handle heavy work in factories as well as delicate tasks like pinching an egg or picking up a potato chip.

A hand without a sense of touch can at best be called a clamp. In a series of smooth moves, how to make the hand touch and feel objects? Sensors act as the nerve cells of the robot and play a crucial role.

From resistors, capacitors, electromagnetics to various flexible electronic skins, the more "nerves" a hand has, the more sensitive it is. Everyone is trying to stuff things into this palm - sized area, making the sensors smaller and more concentrated.

Electronic skin made of nanomaterials

While the hardware is developing, in terms of software drivers, dexterous hands, like their cyber brothers, have also caught the wave of AI. Through various fancy trainings, they have quickly unlocked a bunch of skill points.

For example, at the World Robot Conference we visited recently, there were some hands with slender shapes and flexible movements. Although most of them were exhibited separately, some were installed on robotic arms and could immediately play musical instruments and perform various tasks.

So, the evolution path of dexterous hands is not just their own business.

02

Humanoid robots are developing rapidly. If dexterous hands can't keep up, the overall general upper limit will be restricted.

As there are more and more choices of technological routes, the pricing range and space in the market have also expanded.

We browsed some shopping platforms and found that most robotic hands are priced between a few thousand and tens of thousands of yuan. The cheapest ones may cost only as much as two or three 648 - yuan top - ups in games, while the most expensive ones can cost as much as a Xiaomi SU7.

For example, the following model with an upper - body base but no lower body has a relatively affordable price.

Robotic hands priced at around one or two thousand usually don't have much high - tech, nor are they equipped with high - precision sensors or nice - looking skins. However, it's still okay for them to run some simple algorithms and pose.

However, we can easily tell why cheap things are cheap, but expensive ones have their own reasons. Some are expensive due to the materials, while others are costly because of the motors.

For example, in the following model, not only is the hand itself expensive, but the sensors are even more so. It is said that they are imported high - precision tactile sensors, costing 15,000 yuan each. Installing ten sensors on two hands would cost 150,000 yuan...

During the conversation with Dr. Wang, we found that the pricing difference is indeed significant in different demand scenarios. For example, in pure scientific research, it's quite common for a single hand to cost seven or eight hundred thousand yuan.

These precious little hands have complex technologies and are difficult to mass - produce. They can't be knocked or over - worked during use, otherwise, they'll develop some money - burning problems.

However, if they are used on service robots, the robot body may only cost 30,000 yuan. It would be a bit exaggerated to install a hand worth hundreds of thousands of yuan. Anyway, as long as it can move, there's no need for it to be too precise.

The huge gap between high - end scientific research and practical applications also reflects the complex current situation of the dexterous hand market.

Overall, we feel that it's still difficult to clearly define the pricing of dexterous hands. The raw materials come from all over the place, and it's hard to reduce the cost without mass - production. The measurement standards for each hand are also different, so it's impossible to set a price for the technological content.

For example, some hands' parameters mention sensor accuracy, while others don't. Some hands focus more on the maximum load, while others pay more attention to fingertip perception. One uses domestically developed sensors, while another uses imported motors...

It can only be said that at this stage, each dexterous hand has its own advantages and disadvantages, and which one is better depends on specific needs.

Or maybe when humanoid robots grow up, let them choose for themselves...

03

Finally, everyone may have the same doubt as Jiangjiang - just as robots don't necessarily have to be human - shaped, why do dexterous hands have to pursue five fingers?

Besides the fact that five fingers can do more work in application scenarios, Dr. Wang unexpectedly opened a new world for us from an academic perspective.

Currently, the training data for robots is in short supply, and it all relies on human body collection. Guided by a five - fingered human hand, robotic hands with three or four fingers have a different form and don't learn as well as those with five fingers.

This academic answer makes this anthropomorphic research path seem very correct and reasonable. Although current dexterous hands are not soft, unstable, and even not very dexterous, we are still moving in the right direction.

Maybe we still have a long way to go before we get that legendary hand that can easily open a Coke can, but the idea is clear. Perhaps all that's left is to use technology and time to bridge this long gap.

This article is from the WeChat official account "Chaping X.PIN", author: Jiangjiang. Republished by 36Kr with permission.