The era of AI has come, and I'm starting to take a fresh look at bionic prosthetics.

Over the years, I have been paying attention to the development of the prosthetic industry, especially upper limb prostheses.

To be honest, for a long time in the past, I did not have high expectations for the so-called "intelligent prosthetic hand".

It is not because these products are not well done, but because their technical route determines that they have never really solved a problem:

How to truly replace a person’s lost hand function.

Most of today's myoelectric hands are still essentially "passive prostheses"

Most of the ordinary myoelectric hands on the market today, including many bionic hands priced at hundreds of thousands of yuan, still use the same control logic:

The user sends out myoelectric signals - and the prosthesis performs the corresponding action.

The whole process is:

People give orders and machines execute them.

The machine does not know what is in front of you.

I don’t know what you are going to do.

I don’t even know how hard I should use it.

It just completes the action according to the preset procedure.

If you want to grab an egg, you need to manually control the force;

If you grab a paper cup, you need to manually adjust the mode;

If you want to change the gripping method, you also need to switch to different action modes.

Many times, what is really tiring is not the prosthetic limb, but the user’s brain.

Because of this, upper limb prostheses, especially myoelectric prostheses, have always had a high abandonment rate globally.

The problem is not that the robot hand cannot move, but that it always needs people to constantly "teach" it how to move.

Human hands have never worked by muscles

Many people believe that human hands are controlled by muscles.

Actually not.

When we pick up a water glass, our brain has already completed a series of judgments:

Seeing that it is a water glass;
Determining its position and size;
Estimating whether it is filled with water;
The fingers naturally adjust the gripping posture;
Continuously adjust the strength according to the pressure after contact;
If it is too slippery, it will automatically add force;
If it is too hot, it will release immediately.

We are almost unaware of the whole process.

Because it is not just the muscles that are really involved in the work.

There are also vision, touch, temperature, pressure, experience, judgment, and real-time decision-making of the brain.

This is real manpower.

Why am I now optimistic about bionic prosthetics again?

The answer is only two words:

AI.

To be precise, it is the rapid development of large models, multi-modal AI, and embodied intelligence in recent years.

In the past, prosthetics had only one input - myoelectricity.

Today, we can fully equip prosthetics with:

Myoelectric signal, to understand the patient’s intention of movement;
Camera vision, to understand what objects are in front of them;
Pressure sensor, to know if there is a firm grip;
Temperature sensor, to determine whether there is danger;
The large model is responsible for integrating all information and making real-time decisions.

This means:

Prosthetics no longer just wait for orders.

Instead, start to actively understand the environment.

It can know:

"This is an egg, be lighter."

"This is a thermos cup, hold it tighter."

"This is a hot water kettle. The temperature is too high. Warning the user."

"This object begins to slip, and the grip needs to be increased."

This is no longer myoelectric control in the traditional sense.

It is an active prosthesis in the true sense.

From “controlling machines” to “machine understanding people”

In the past few decades, the entire prosthetic industry has been working hard to improve control accuracy.

For example:

Myoelectric recognition is more accurate;

More movements;

Motor speed is faster;

More fingers.

These are all important.

But I think they still only solve the problem of "execution".

What is really missing has always been:

Understanding.

When should the machine take the initiative?

When should you wait?

When should users be protected?

When should automatic corrections be made?

These are not problems that traditional algorithms are good at solving.

This is exactly what AI does best.

AI makes active prostheses truly possible for the first time

A few years ago, if someone told me:

False The limb can understand the environment on its own;

can actively determine the grasping method;

can continuously learn user habits;

can even become more and more like a human hand.

I most likely think this is just a concept.

But today, I don’t think so.

Because of the development of open source large models, multi-modal perception, edge computing, and low-power AI chips, these technologies have begun to have a realistic basis.

Of course, there is still a long way to go before it is truly mature.

Cost, power consumption, real-time performance, security, regulatory certification, and clinical verification are all challenges.

But at least, this technical route is clearly visible.

Personal opinion

I personally believe that the real competition in the prosthetic industry in the future is no longer just:

Whose motor is better;

Whose mechanical structure is more advanced;

Whose electromyographic recognition accuracy is higher.

But who can be the first to create a truly active prosthetic limb.

Really excellent prosthetics should not just execute people’s orders.

It should be able to understand people's intentions, understand the environment, and actively help users complete every action in life.

This may be the direction in which prosthetics should really develop in the AI era.

This article represents only the author’s personal views.

I do not deny the value of existing myoelectric prostheses and bionic hands in the industry. They have promoted the development of prosthetic technology and have also helped many patients.

This article just shares some of my thoughts on the development direction of prosthetics in the AI era. I believe that different technical routes have their own value. I also welcome R&D personnel, engineers, clinical experts in the industry, and all friends who are concerned about the development of prosthetics to leave messages and exchange messages to discuss the future development possibilities of active intelligent prostheses.

I predict that whoever makes active prosthetics well in the future will depend on who collects more real user data and whether they have their own large data model.

The era of AI has come, and I&#039;m starting to take a fresh look at bionic prosthetics.

The era of AI has come, and I'm starting to take a fresh look at bionic prosthetics.